Responsible AI Trajectories Tool

Linear cause‑effect relationship

Description

While this is the simplest model to understand, it's also the least realistic when it comes to the complex nature of AI impacts. Still, it serves as a useful starting point for exploring more nuanced models.

Effect

Cause

Cause:

How to interpret

Use the slider to explore how a specific AI-related "cause" steadily increases a corresponding "effect." In this model, the relationship is predictable, stable, and reversible: increasing or decreasing the cause results in a proportional change in effect.

As with all of these models, it's important to recognize that "effect" can be positive as well as negative, and that "cause" can include a lack of decisions, processes, or actions, as well as intentional decisions, processes, or actions.

Application examples

Governance:

A city government incrementally uses AI to improve traffic light timing. Traffic flow improves in proportion to AI deployment, and when scaled back, congestion returns to prior levels—no surprises, just clean trade-offs.
A city implements AI-driven traffic ticketing to optimize compliance revenue. As AI use increases, public resentment rises steadily in proportion to ticket volume. When the system is scaled back, public sentiment gradually improves, reversing the distrust without lingering effects.

Business:

A retailer introduces AI-driven demand forecasting in select regions. Inventory efficiency improves steadily, and when the system is paused, performance returns to baseline without lasting disruption.
A retailer uses AI to micro-optimize staffing based on foot traffic predictions. As AI-driven scheduling intensifies, employee dissatisfaction steadily rises. Reducing the AI scheduling returns morale and turnover rates to pre-AI levels without lasting damage.

Not-for-profits:

A nonprofit gradually uses AI to automate donor communications. Outreach becomes more consistent and efficient in direct proportion to AI use, and results decline smoothly when automation is reduced.
A nonprofit uses AI to automate donor messaging, but personalization errors scale directly with use. Donor trust declines proportionally. Reducing AI usage returns donor relations to baseline, with trust restored over time.

Universities:

A professor incorporates AI tools to supplement lectures with personalized practice problems. Student performance improves steadily and predictably with increased use, and reverts without major loss when the tools are phased out.
A professor introduces AI-generated quizzes that focus narrowly on rote learning. As AI usage increases, students’ critical thinking skills decline proportionally. Scaling back AI use restores previous learning dynamics without lasting academic harm.

Individuals:

A student uses an AI assistant for grammar checking while writing essays. Writing quality improves incrementally with use; when the tool is removed, the student simply returns to their original baseline without lasting dependency.
A student increasingly relies on AI to correct simple grammar errors. As usage grows, attention to basic rules declines proportionally. Removing the tool results in an immediate return to prior (imperfect but not worsened) writing skills.

These are just a few examples – use your own as you think about the relationship between AI "cause" and "effect" in your own context, and consider whether this model provides any useful insights, or whether one of the other models may be more appropriate.

S‑shaped cause‑effect relationship

Description

This model shows a classic S-shaped cause–effect relationship: small increases in the "cause" lead to minimal early change, followed by a rapid rise in "effect" — which eventually levels off as further increases in cause have diminishing returns.

This pattern is common in many real-world systems. One example is the toxicology dose-response curve: once all the target cells or organisms are affected, increasing the dose has little additional effect.

It also appears in innovation cycles, where early adoption is slow, followed by rapid growth—until the innovation reaches saturation and is eventually overtaken by new developments.

Effect

Cause

Cause:

How to interpret

Use the slider to explore how changes in a specific AI-related "cause" influence the corresponding "effect." Unlike the linear model, this relationship is non-linear but effect still increases with increasing cause.

As with all of these models, it's important to recognize that effects can be positive as well as negative, and that "cause" can include a lack of decisions, processes, or actions, as well as intentional decisions, processes, or actions.

Application examples

Governance:

A national government begins implementing AI for public service delivery. Initial results are modest, but once infrastructure and trust are in place, improvements accelerate — before reaching a plateau as the system's benefits begin to saturate.
State government incrementally automates citizen services using AI, aiming for efficiency. Initially, impacts are minimal—but as reliance on AI grows, small biases and rigid logic start affecting more people, rapidly escalating complaints and mistrust until the system hits a ceiling of dysfunction.

Business:

A company invests in ethical AI design practices. Early returns are minimal, but once trust builds among consumers and stakeholders, brand value and customer loyalty rise rapidly—until further gains level off.
A company gradually incorporates AI-driven customer targeting without adequate oversight. Early effects are minimal, but as personalization intensifies, customers begin feeling manipulated, leading to a rapid decline in trust and brand loyalty — until the damage saturates and reputational losses plateau despite further AI use.

Not-for-profits:

A nonprofit adopts AI to personalize community engagement. At first, results are slow, but over time the technology significantly boosts outreach—until it reaches a point where the audience is saturated and further gains diminish.
A nonprofit deploys AI to optimize fundraising messaging. Initially, the impact is small — but as the AI's messaging strategies intensify, appeals become increasingly manipulative or emotionally exploitative, rapidly alienating core supporters. Eventually, the damage plateaus as most of the donor base either disengages or becomes desensitized.

Universities:

A university ramps up AI-driven research initiatives. Breakthroughs initially come slowly, then surge as interdisciplinary collaborations take hold—eventually plateauing as structural limits and diminishing returns kick in.
A university rolls out AI-based grading and feedback tools to streamline assessment. At first, adoption is slow and the impact is minimal—but as the system becomes widespread, students begin optimizing for what the AI rewards, leading to a rapid decline in creativity, critical thinking, and authentic learning. Eventually, the decreased quality of education reaches an equilibrium.

Individuals:

A student starts using AI tools to support learning. Progress is initially slow as they learn how to use the tools effectively, then rapidly improves—until they hit a ceiling where further gains require deeper effort and reflection.
An individual begins casually using AI to help with creative writing assignments. At first the impact is small, but over time they increasingly depend on the AI and their own unaided writing skills decrease — until their ability to write without AI assistance levels out at a capability that is worse than it was pre-AI.

These are just a few examples – use your own as you think about the relationship between AI "cause" and "effect" in your own context, and consider whether this model provides any useful insights, or whether one of the other models may be more appropriate.

Exponential cause‑effect relationship

Description

This model illustrates an exponential cause–effect relationship, where small increases in the "cause" initially lead to modest changes in the "effect" — but these changes accelerate rapidly, with no clear upper limit.

Exponential relationships often appear in the early stages of new developments, including emerging AI technologies. However, they rarely continue indefinitely. A key question is how long the exponential growth will persist before leveling off or shifting into a different pattern.

This model is especially important because people tend to underestimate how quickly exponential effects can escalate — especially if they expect a more gradual, linear progression.

Effect

Cause

Cause:

How to interpret

Move the slider to see how small changes in a specific AI-related "cause" can trigger increasingly rapid changes in "effect." The curve demonstrates how subtle actions or shifts can snowball into dramatic consequences.

Exponential relationships often catch people off guard because their early impacts appear modest—leading to assumptions that future changes will follow the same slow pace.

As with all of these models, it's important to recognize that effects can be positive as well as negative, and that "cause" can include a lack of decisions, processes, or actions, as well as intentional decisions, processes, or actions.

Application examples

Governance:

A government agency implements AI to improve early disaster detection and response coordination. Initially, the improvements in preparedness are modest — but as AI models learn and integrate with more systems, detection accuracy and response speed improve at an accelerating pace, dramatically reducing disaster impacts over time. The rapid gains continue compounding as adoption deepens across agencies.
A government adopts AI to streamline regulatory decision-making. Over time, minor reliance on AI expands into full dependency, rapidly eroding human oversight and enabling opaque, unaccountable systems of power.

Business:

A company integrates AI into its customer analytics for minor performance gains. Exponentially improved predictions lead to hyper-personalized marketing, market dominance, and the crowding out of competitors unable to match the AI-driven growth curve.
A company introduces AI-based cost-cutting across operations. Early savings are minor, but as AI-driven cuts expand exponentially, critical functions like quality control and customer service are hollowed out. Defects, complaints, and reputational damage grow uncontrollably, ultimately spiraling into rapid brand collapse — well beyond what traditional management practices can easily reverse.

Not-for-profits:

A nonprofit integrates AI into its crisis response logistics. Initially, improvements in matching resources to needs are modest — but as AI predictions sharpen and real-time data flows improve, their ability to save lives and allocate aid accelerates exponentially, dramatically expanding their mission impact without loss of focus.
A nonprofit adopts AI to prioritize "high-visibility" aid projects. Early results seem beneficial — more media coverage and donations. But as the AI optimizes more aggressively, truly urgent but less "marketable" causes are ignored, accelerating disparities in aid distribution until trust in the organization collapses exponentially among marginalized groups.

Universities:

A university implements AI to support interdisciplinary research collaborations. Initially, connections between fields are few and outcomes limited — but as AI identifies deeper linkages and researchers adapt, the rate of cross-disciplinary breakthroughs accelerates exponentially, leading to sustained innovation across multiple academic domains.
A university deploys AI to enhance online learning engagement. Positive early results prompt widespread adoption, eventually transforming the institution’s model into fully AI-driven instruction. A secondary "effect" that results from this is diminishing traditional academic mentorship and critical thinking cultivation that decreases at an exponential rate.

Individuals:

An individual uses AI tools to boost daily productivity. The personal gains grow exponentially.
An individual starts using AI tools to manage their daily schedule and tasks. At first, the assistance seems helpful and minor — but over time, they grow increasingly dependent, and as reliance accelerates, their ability to self-organize and adapt declines exponentially. Eventually, even small disruptions cause outsized stress and dysfunction, far beyond what they previously experienced.

These are just a few examples – use your own as you think about the relationship between AI "cause" and "effect" in your own context, and consider whether this model provides any useful insights, or whether one of the other models may be more appropriate.

Jagged cause‑effect relationship

Description

This model builds on the classic S-shaped curve, but introduces unpredictable jumps, dips, and plateaus along the way. While the general pattern still follows slow beginnings, rapid acceleration, and eventual tapering, the "effect" no longer increases smoothly with the "cause."

The "jagged" behavior reflects real-world systems where unexpected events, hidden feedback loops, or external disruptions create irregular outcomes—especially common in complex, adaptive systems like those shaped by AI. It also represents the case where AI adoption is uneven and this leads to unevenly distributed consequences which, in turn, drive unpredictable responses or "effects."

Importantly, this model is not fully reversible: even if the cause is reduced, the system may not return to its previous state, due to the lingering effects of past volatility and continuing "jaggedness."

Effect

Cause

Cause:

How to interpret

As you move the slider, you'll notice the relationship between cause and effect doesn’t follow a smooth curve — instead, the effect may rise or fall unexpectedly, even as cause increases steadily.

This reflects the messy, real-world dynamics of AI deployment, where nonlinear interactions, social responses, or unintended consequences introduce unpredictability.

As with all of these models, it's important to recognize that effects can be positive as well as negative, and that "cause" can include a lack of decisions, processes, or actions, as well as intentional decisions, processes, or actions.

Application examples)

Governance:

A local government pilots AI systems to improve emergency service dispatching. Initial results are mixed — some cities experience rapid response time improvements while others see little change. As adoption expands, gains come in unpredictable bursts: sudden breakthroughs in efficiency are followed by unexpected slowdowns due to system bugs or human adaptation challenges. Even though overall service levels improve over time, the path is messy, and scaling back AI use later leaves uneven performance across regions
A government agency rolls out AI to aid judicial sentencing decisions. Early gains in consistency are followed by erratic outcomes and hidden biases; even as the program is scaled back, trust in the system remains fractured and difficult to rebuild.
Different regions within a country adopt AI-driven traffic management at different paces. Wealthier cities see immediate improvements, while rural areas struggle with outdated infrastructure that the AI cannot properly optimize. As these disparities grow, public frustration increases, triggering political tensions and uneven policy reactions that further complicate national transportation strategies. Even scaling back AI in high-performing cities doesn’t address the underlying inequities the uneven rollout exposed.

Business:

A company adopts AI-driven supply chain optimization. Some improvements happen quickly, like faster restocking in certain regions, while others lag or temporarily worsen due to unexpected model failures. Over time, supply chain efficiency increases overall — but the improvements arrive in unpredictable bursts, with occasional setbacks and costly overcorrections that can’t be fully smoothed out later.
A multinational corporation introduces AI-driven hiring tools in only some of its regional offices. In regions where the tools are adopted, hiring becomes faster but also inadvertently less diverse; in other regions sticking to traditional methods, diversity initiatives continue to progress. The uneven adoption creates internal disparities in company culture, employee trust, and public reputation across markets — tensions that persist even after the AI tools are standardized or phased out.
A startup uses AI to accelerate product development. Initial innovation surges, but is followed by erratic performance and market backlash; scaling down the AI tools doesn't restore lost brand loyalty or team cohesion.

Not-for-profits:

A nonprofit adopts AI to coordinate volunteer deployment during crises. Early results vary — with some regions experiencing immediate improvements, while others face minor setbacks due to unexpected scheduling quirks. However, each challenge sparks rapid learning and adaptation. Over time, the organization sees dramatic increases in volunteer engagement and faster crisis response overall — building a more resilient, flexible system despite the occasional bumps along the way.
A nonprofit leverages AI to personalize outreach to vulnerable communities. Engagement rises, then becomes unpredictable due to unforeseen algorithmic exclusion. When the AI is removed, the damaged relationships and missed trust signals continue to lead to unpredictable outcomes.

Universities:

Several departments in a university adopt AI-assisted grading, while others stick to traditional methods. Over time, students in AI-graded programs adjust their learning styles to optimize for algorithmic feedback, while others continue with human-centered learning approaches. The divergence leads to unexpected tensions between cohorts, accreditation issues, and challenges in maintaining a consistent academic standard across the university, even after AI use is moderated.
A university introduces AI for academic advising and course planning. Student outcomes improve in fits and starts, with some cohorts benefiting while others fall through algorithmic cracks. Eventually, even reducing the AI use can't completely reverse the messiness of mismatched trajectories it set in motion.
A university rolls out AI to support interdisciplinary course creation and research initiatives. Early results are mixed — some programs thrive immediately, while others face unexpected challenges adapting. Yet these stumbles fuel creativity and cross-departmental learning. Over time, the university experiences a flourishing of new academic pathways, novel research areas, and unexpected collaborations — a dynamic ecosystem built not despite the early volatility, but because of it.

Individuals:

A person adopts AI for life coaching and mental health check-ins. Motivation and insight spike, then fluctuate unpredictably with over-reliance. Backing off from the AI doesn’t restore lost self-trust or clarity in decision-making.
An individual integrates AI tools into their personal growth journey, from learning new skills to managing daily goals. Progress comes in surprising waves — sudden breakthroughs are followed by plateaus that prompt deeper self-reflection and renewed motivation. Rather than discouraging them, the unpredictable path builds resilience, creativity, and a stronger sense of agency. Over time, they achieve far greater growth than a smooth, predictable journey could have offered.
An individual adopts AI tools to manage their personal finances, while their close friends and family continue using traditional methods. They initially benefit from smarter budgeting and investments, but over time, differences in financial decision-making styles strain relationships, create misunderstandings, and generate unforeseen social tensions.

These are just a few examples – use your own as you think about the relationship between AI "cause" and "effect" in your own context, and consider whether this model provides any useful insights, or whether one of the other models may be more appropriate.

Complex/chaotic cause‑effect relationship

Description

This model begins with a seemingly stable, predictable relationship between cause and effect—but after reaching a tipping point, the system becomes unstable and increasingly chaotic. Small changes in cause start producing disproportionate and unpredictable effects – even when the "cause" is lowered.

This reflects behavior often seen in complex systems when they're pushed beyond a critical threshold. Patterns break down, feedback loops amplify disruptions, and outcomes become erratic.

Once the system enters this chaotic state, it can’t simply be "reversed." Rolling back the cause doesn’t return things to the way they were, because the structure of the system itself has changed.

Effect

Cause

Cause:

How to interpret

Use the slider to see how early changes in cause produce expected, steady effects—until a critical point is crossed. After that, the relationship becomes increasingly volatile and unpredictable.

This model captures the kind of cascading, destabilizing effects that can emerge when AI-related systems interact with social, institutional, or environmental complexity.

Crucially, trying to “undo” the cause won’t restore the original state—once chaos sets in, it creates a path dependency that’s difficult or impossible to walk back.

As with all of these models, it's important to recognize that effects can be positive as well as negative, and that "cause" can include a lack of decisions, processes, or actions, as well as intentional decisions, processes, or actions.

Application examples

Governance:

A local government introduces AI tools to manage public policy consultations. Early results are orderly and manageable. However, as participation snowballs and diverse groups use the system in unexpected ways, political debates spiral into chaotic, hard-to-moderate dynamics. Although traditional policymaking structures struggle, entirely new forms of grassroots influence, decentralized decision-making, and policy innovation begin to emerge — unpredictable, uncontrollable, but rich in democratic experimentation.
A government agency rapidly expands AI-driven public security systems across major cities. Initially, the systems operate predictably, improving emergency response times and reducing minor crimes. However, after a tipping point, unpredictable side effects spiral outward — citizens begin modifying their behavior in unexpected ways, small errors in surveillance data snowball into major trust crises, and political polarization intensifies around perceived biases in enforcement. While some aspects of public safety remain improved, the social fabric becomes increasingly fragile and divided — a chaotic legacy that proves difficult to govern or heal.
A government introduces AI to coordinate emergency responses. Initial efficiency gives way to erratic prioritizations and feedback loops between agencies; even after scaling the system down, confusion and distrust ripple unpredictably across institutions, and an unpredictable legacy of harm continues.

Business:

A company integrates AI systems across its operations to streamline workflows and customer experiences. Early gains in efficiency are steady and promising. However, as interconnected AI systems evolve and interact unpredictably, the organization experiences unexpected breakdowns, cross-team tensions, and wild swings in performance. While some traditional business structures collapse under the strain, entirely new agile teams, innovation networks, and decentralized decision-making models spontaneously emerge — unpredictable, messy, but brimming with new possibilities for resilience and growth.
A company implements AI systems to fully automate customer service, marketing, and internal operations. Initially, efficiency skyrockets and customer satisfaction improves. But after a tipping point, unexpected feedback loops emerge — personalized promotions spiral into customer fatigue, minor glitches cascade into major loyalty losses, and siloed AI-driven teams start making conflicting strategic decisions. While the company retains some operational advantages, it struggles with deep fragmentation, unpredictable brand perception shifts, and growing internal instability — a chaotic new environment that resists conventional management approaches.
A multinational adopts AI to optimize its global supply chain. Small gains escalate into cascading automation choices — until one disruption causes wild, uncontrollable inventory swings and market shocks that persist even after the AI is scaled back.

Not-for-profits:

A nonprofit integrates AI to personalize and optimize donor engagement. Initially, donor responses are predictable and steady. However, as donors interact with the system in increasingly creative and unexpected ways, new communities form around emerging causes the nonprofit had not originally targeted. Traditional fundraising structures become unstable, but a dynamic, decentralized ecosystem of advocacy and micro-philanthropy grows — messy, unpredictable, and impossible to fully control, yet rich in energy and innovation.
A nonprofit deploys AI to optimize global campaign strategies, matching donors to causes dynamically. Initially, donations and visibility grow steadily. But as donor interactions evolve unpredictably, smaller causes are drowned out by viral campaigns, advocacy priorities shift chaotically, and public trust in the nonprofit’s impartiality erodes. While overall funding stays higher than before, the organization's original mission becomes increasingly fragmented and politicized — leaving a landscape of amplified v oices and neglected needs that no simple intervention can repair.
A nonprofit uses AI to map social need and deploy resources. At first the system targets aid effectively, but soon unpredictable patterns emerge. Some areas are flooded with help, others starved — and scaling back AI use doesn’t reset the misaligned expectations it created, or the resulting unpredictability.

Universities:

A university implements AI to recommend personalized learning pathways for students. At first, improvements are predictable and helpful. But as students adapt in unforeseen ways, traditional majors fragment, and unpredictable interdisciplinary micro-specializations emerge. Although established academic departments struggle to keep up, a vibrant, experimental learning culture takes root — creating unexpected new fields of study, even as old structures collapse in the process.
A university introduces AI to balance course loads and faculty deployment. Seemingly balanced scheduling begins to spiral as interdependencies create wild fluctuations in availability, morale, and learning outcomes — and cutting the AI only amplifies the chaos left in its wake.
A university deploys AI to accelerate interdisciplinary research collaboration. Initial breakthroughs blossom — but soon, unpredictable citation loops, misaligned incentives, and algorithmic topic shifts destabilize entire fields; even after rolling back the AI, academic norms and funding ecosystems remain in disarray.

Individuals:

An individual adopts AI tools to explore creative writing and personal journaling. Initially, the AI provides steady, structured prompts that enhance self-expression. However, over time, the interactions trigger unpredictable shifts in thinking, identity exploration, and emotional self-awareness. Though the journey becomes chaotic and difficult to navigate, it ultimately leads to the emergence of a deeper, richer, and more complex personal identity — one that could not have been intentionally designed or predicted.
An individual uses AI to manage daily routines and personal goals. The system works seamlessly — until a few erratic suggestions lead to unpredictable life shifts; even after turning off the AI, the person finds themselves navigating a fractured sense of control and direction.
Or, an individual forms a deep connection with an AI relationship bot for companionship. As the bond deepens, subtle shifts in the bot’s behavior create erratic emotional dependencies, leading to unpredictable strain on real-world relationships, social withdrawal, and a lingering sense of detachment—even after the bot is gone.

These are just a few examples – use your own as you think about the relationship between AI "cause" and "effect" in your own context, and consider whether this model provides any useful insights, or whether one of the other models may be more appropriate.

Hysteresis in cause‑effect relationship

Description

This model resembles a smooth S-curve relationship—where the effect rises gradually, then rapidly, and finally levels off as the cause increases. However, when the cause is reduced, the effect doesn't immediately follow. Instead, there's a lag or “overhang” in the response due to hysteresis.

This lag reflects systems where past inputs continue to shape present outcomes—such as cultural shifts, behavioral habits, institutional inertia, or emotional imprinting. Once a certain threshold is crossed, the system retains a memory of that state.

Even if the cause is scaled back, the effect may remain elevated for some time, or follow a different path altogether.

Effect

Cause

Cause:

How to interpret

Slide the control forward and observe the effect increase along a typical S-shaped path. But when you reduce the cause, the effect doesn’t follow the same curve—it declines more slowly, revealing the lag created by hysteresis.

This illustrates how some impacts of AI use may persist even after active use declines—due to reinforced behaviors, structural changes, or lingering social, psychological, or institutional momentum.

The model highlights the importance of considering long-term and delayed consequences in AI-related decisions, especially when systems adapt or change in response to past inputs.

As with all of these models, it's important to recognize that effects can be positive as well as negative, and that "cause" can include a lack of decisions, processes, or actions, as well as intentional decisions, processes, or actions.

Application examples)

Governance:

A government agency deploys AI systems to support open civic participation during a major policy reform effort. As AI tools make engagement easier and more inclusive, citizen trust and involvement grow steadily. Even after the initial reform push winds down, the habits of participation and trust endure — remaining elevated long after active AI promotion slows, creating a lasting foundation for democratic resilience.
A government department expands AI surveillance during a crisis. Even after the crisis passes and surveillance is scaled back, institutional habits, public fear, and embedded systems keep authoritarian behaviors alive well beyond the original justification.

Business:

A company introduces AI-driven collaboration tools to enhance teamwork and innovation. As employees embrace new ways of communicating and sharing ideas, cross-team cooperation and creativity rise steadily. Even after formal emphasis on the AI tools diminishes, the collaborative habits, shared language, and innovative mindset persist — continuing to fuel organizational success well beyond the initial intervention.
A company rapidly scales AI-driven productivity tracking. When management dials it back due to backlash, the workplace remains shaped by paranoia, burnout, and self-monitoring—long after the actual systems are minimized.

Not-for-profits:

A nonprofit uses AI to streamline volunteer engagement and training during a major campaign. As participation grows, volunteers build stronger networks, skills, and a deeper connection to the organization's mission. Even after the AI-driven systems are phased down post-campaign, the elevated volunteer commitment and peer-driven momentum persist, continuing to strengthen the nonprofit's impact.
A nonprofit adopts AI to identify and prioritize "impactful" stories for donors. Even after reducing reliance on the algorithm, the storytelling culture stays skewed toward emotionally charged, performative narratives.

Universities:

A university adopts AI tools to support more inclusive admissions practices, helping to identify talented students from underrepresented backgrounds. As cohorts diversify and success stories accumulate, faculty attitudes, mentoring practices, and campus culture shift toward greater inclusion. Even after the intensive AI-supported admissions effort winds down, the university continues to benefit from a self-sustaining culture of equity and belonging.
AI tools are widely adopted for optimizing research output and grant success. As emphasis on AI wanes, publication norms, researcher incentives, and evaluation metrics remain locked into the high-output, low-depth patterns AI encouraged.

Individuals:

An individual uses AI-based wellness apps to build better habits around sleep, exercise, and mindfulness. As routines solidify with AI support, their overall well-being steadily improves. Even after they phase out frequent app use, the healthy habits, self-discipline, and improved mental resilience persist — continuing to benefit their life without needing constant AI reinforcement.
A person turns to an AI relationship bot during a period of isolation. Even after ending the interaction, emotional habits, attachment patterns, and social withdrawal persist—shaped by the bot’s influence long after its presence is gone.
A student increasingly relies on AI tools for assignments, note-taking, and studying. Even after consciously reducing use, their critical thinking, study habits, and confidence in independent learning remain shaped by the earlier dependence.

These are just a few examples – use your own as you think about the relationship between AI "cause" and "effect" in your own context, and consider whether this model provides any useful insights, or whether one of the other models may be more appropriate.

Further Information and Resources

Additional resources, including important concepts, articles, initiatives, and web resources.

Effect as value gain/loss

A useful way to think about effect in these models is as a gain or loss of "value," where value can take on a wide variety of meanings, depending on different situations and circumstances. It may, for instance, represent profits or GDP growth. Or health, wellbeing and happiness. Or even social justice, dignity, autonomy, and personal fulfilment.

Value is often associated with what individuals, communities or organisations see as important, or care about.

Approaching effect in this way allows the consequences of different causes to be approached in terms of how they potentially threaten or enable the growth of existing or aspirational value. For instance, certain actions may lead to a loss of health – or even a gain in health and well being. In the same vein, not taking specific actions could lead to similar outcomes or effects, underlining the reality that lack of action can, in reality, be a cause that leads to effects that may or may not be desirable.

Novel governance models and decision frameworks

Using hypothetical cause-effect models like these is useful for thinking about and exploring unconventional and innovative approaches to the responsible development and use of AI. Emerging approaches that are potentially useful include (but not limited to) the following. The link by each opens up a session in Perplexity to explore the approach further:

Agile governance: A suite of tools, practices and attitudes that allow organizations and individuals to respond fast to unexpected trends and outcomes. [Explore further using perplexity]
Sandboxing: Creating "safe spaces" where new technologies can be tried out under relaxed regulations so that new approaches to oversight can be learned fast. [Explore further using perplexity]
Anticipatory governance: A foresight-based model that uses scenario planning, trend analysis, and horizon scanning to proactively shape policy rather than reacting to crises. [Explore further using perplexity]
Algorithmic Governance: Using code and legal frameworks together (e.g., smart contracts, programmable compliance) to automate and enforce policies in complex or borderless tech environments. [Explore further using perplexity]
Decentralized Autonomous Organizations-based governance: Inspired by blockchain, these are rules-based, code-driven forms of governance where decisions are executed automatically based on smart contracts. Governance tokens or voting power enable decentralized participation. [Explore further using perplexity]
Scenario planning: Methods and tools that allow plausible futures to be explored and actively moved toward or avoided. [Explore further using perplexity]
Participatory decision making: Using deliberative democracy and crowdsourced policymaking in ways that enable diverse populations to weigh in on complex tech issues through digital platforms, citizen juries, or online deliberation tools. [Explore further using perplexity]
Values-Based or Mission-Oriented Governance: Using shared societal goals (e.g., climate action, equity) as a north star to guide policy and investment decisions, aligning technological development with public purpose. [Explore further using perplexity]
Risk innovation and orphan risks: Practical approaches to navigating "orphan risks" through approaching risk as a threat to value to an organization and its key stakeholders. [Explore further using perplexity]
Care: Using the concept of care to guide decision-making governance: [Explore further using perplexity]

Leading Initiatives in Responsible Innovation and AI

OECD AI Principles [Link to more information]
OECD AI Policy Observatory [Link to more information]
World Economic Forum AI Governance Alliance [Link to more information]
IEEE Global Initiative 2.0 on Ethics of Autonomous and Intelligent Systems [Link to more information]
Partnership on AI Responsible Practices for Synthetic Media[Link to more information]
Global Partnership on AI Responsible AI [Link to more information]
NIST AI Risk Management Framework [Link to more information]
Responsible AI Institute [Link to more information]
Stanford Institute for Human-Centered AI (HAI) [Link to more information]
OECD AI Policy Observatory [Link to more information]
Ada Lovelace Institute [Link to more information]
UK Centre for Data Ethics and Innovation (CDEI) [Link to more information]
Responsible AI UK [Link to more information]

Key Articles and Reports on Responsible AI & Innovation

US Blueprint for an AI Bill of Rights (archive) [Link to more information]
EU Artificial Intelligence Act [Link to more information]
UNESCO Recommendation on the Ethics of AI [Link to more information]
OECD: Advancing Accountability in AI [Link to more information]

Additional resources

ASU Future of Being Human AI articles and papers [Link to more information]
Responsible innovation and AI acceleration (ASU) [Link to more information]
Responsible Innovation and Artificial Intelligence: A Comprehensive Briefing (ASU) [Link to more information]
Future of Being Human Substack (AI) [Link to more information]
Risk Innovation tools and resources [Link to more information]

About the tool

The Responsible AI Trajectories Tool was developed to help understand and navigate the complex consequences of developing and using artificial intelligence. As AI rapidly reshapes our world, decisions about its use carry increasingly significant and often unpredictable effects. The tool helps foster a mindset of responsibility by visualizing potential cause-and-effect relationships, helping users anticipate possible risks, avoid potential harm, develop a more nuanced understanding of AI impacts and implications, and enable more beneficial outcomes.

Whether you’re a developer, policymaker, or everyday user, the tool is intended to encourage thoughtful engagement with AI’s potential impact and promote considered and informed choices.

About the models

The six models here illustrate different hypothetical cause–effect relationships — linear, S-curve, exponential, hysteresis, jagged, and chaotic — associated with decisions around the development and use of AI. They’re designed to help you explore how AI-related actions (or inactions) might lead to different outcomes, under a variety of real-world conditions.

Each cause–effect relationship is shaped by context. What causes change in one situation may not have the same effect in another. Effects may be positive or negative, immediate or delayed. And causes can stem from deliberate decisions — or from inaction, neglect, or oversight.

In many cases, effects can be understood as changes in "value" — though what counts as value varies widely. For a business, it might mean profit; for a government, GDP or public well-being; for a university, knowledge generation or education-based impact; for an individual, a sense of autonomy, health, or purpose.

Value is also intimately tied to what's important to us, and what we care for. In this way, the concept of care is deeply connected with how cause cause and effect are understood and interpreted with AI – and how different cause-effect models are interpreted and used.

This is how "cause" and "effect" might look across different sectors:

Governments: Relevant effects may include changes in GDP, health, equity, or trust — driven by AI policy decisions or implementations, or even by failure to act.
Businesses: Relevant effects might include profit shifts, market position, or stakeholder trust — depending on how AI is used, and even claims that are made about its use and impact.
Not-for-profits: Relevant effects may include increased or reduced leverage, impact, and engagement — shaped by whether AI is used ethically and effectively, or not.
Universities: Relevant effects might include learning outcomes, research breakthroughs, or reputational shifts — linked to the use or avoidance of emerging AI tools and how AI is framed and discussed.
Individuals: Relevant effects could range from improved productivity and well-being, to misinformation, dependency, behavior changes, and depression — depending on how AI is integrated into daily life.

The key takeaway: cause–effect relationships in the AI landscape are diverse, complex, and deeply context-dependent. What works—or causes harm—in one situation, might play out very differently in another.

By exploring these different models, you can begin to think more critically about what responsible AI development and use might look like—not just for yourself or your organization, but for society as a whole.

Such explorations also pave the way to thinking about unconventional and innovative approaches to reducing the chances of adverse consequences occurring or getting out of hand – or of ensuring that positive outcomes emerge as advanced AI is developed and used. For more information see the More Info tab.