An atlas of better possibilities
Products, cures, and insights that could turn increasingly capable AI into a better everyday life.
What could abundance actually look like? Explore concrete positive scenarios enabled by AI—from useful products and possible treatments to new scientific understanding. These are grounded possibilities, not promises or dated forecasts.
These are positive possibilities, not promises or timeline forecasts. Lists put the most intriguing ideas first: an editorial judgment about interest and transformative potential, not likelihood or arrival date. The labels describe what must happen next, not numerical probabilities.
Products are useful tools and services; cures include potential treatments and prevention, not claims of proven cures; insights are discoveries that help us understand the world.
6 capabilities · 30 positive scenarios · Most intriguing first in each list
30 scenarios to explore
Better ideas, found by calculation
Faster algorithms, better engineering, and new scientific insights—if AI turns hard mathematics into usable methods.
From biological trial and error to guided discovery
Better treatment candidates, more precise experiments, and eventually predictive models of how living cells respond.
A small team, a much larger canvas
More people can explain, teach, visualize, and tell stories without a traditional studio-sized production budget.
Interactive worlds for more than entertainment
Smaller creators build richer games, and teachers and trainers gain new ways to let people learn by doing.
Search for the substance a better world needs
New batteries, catalysts, membranes, and electronic materials become easier to discover and improve.
Abundance that can pick things up
AI moves beyond digital output to help with routine physical work, scientific experiments, and hazardous tasks.
Solving math · Product
Plausible next step
Engineers specify loads, materials, cost, and performance; software proposes highly efficient components worth manufacturing.
Engineering design often involves choosing among millions of combinations of dimensions, materials, and operating parameters. AI-assisted optimization could help explore those combinations while a physics model checks constraints such as stress, temperature, or vibration.
The output is most valuable when an engineer can inspect the tradeoffs rather than accept one mysterious 'best' answer. A lightweight component that is impossible to manufacture, difficult to repair, or sensitive to small defects may be worse than a simpler design.
A manufacturer searches for a lighter motor bracket that carries the same load. The system proposes several shapes with different manufacturing and maintenance tradeoffs, and engineers test the finalists.
Physical models, manufacturability, and real-world testing still determine whether a design works.
Background, not proof this scenario is available.
Solving math · Product
Breakthrough-dependent
Aircraft, turbines, and ventilation systems become cheaper to develop and more efficient to operate.
Turbulent flow contains interacting structures at many scales. Resolving every relevant eddy can be prohibitively expensive, so practical simulations use approximations. New mathematics and AI-assisted modeling could improve those approximations or concentrate computation where it matters.
A useful system would predict not just a best estimate but also when its confidence is low. Engineers still need physical measurements to check whether the model works for a new aircraft shape, operating regime, or unusual weather condition.
A designer compares several turbine-blade shapes virtually, then uses a smaller set of physical tests to confirm efficiency and loading at different operating conditions.
The Navier–Stokes existence-and-smoothness problem is not itself a solution to turbulence. Chaotic prediction limits remain.
Background, not proof this scenario is available.
Solving math · Insight
Plausible next step
A relationship discovered in geometry or algebra becomes a reusable tool across several disciplines.
AI can compare large collections of mathematical objects and identify properties that seem connected. Such patterns can suggest a conjecture that a mathematician would not have thought to test.
The valuable endpoint is more than a statistical association: it is an explanation, a theorem, or a reusable method. Researchers must search for counterexamples and establish where a proposed relationship does and does not apply.
An unexpected relationship between two measurements of a mathematical structure leads researchers to a new conjecture. A proof explains the relationship and makes it useful beyond the original dataset.
Interesting patterns are not proofs, and a valid proof may have no immediate engineering application.
Background, not proof this scenario is available.
Solving math · Product
Plausible next step
Critical controllers arrive with machine-checkable evidence that they meet specific requirements.
Formal verification translates a precise requirement into a mathematical statement about a program. AI can help write the specification, propose a proof, or find an input that violates it; a proof checker then verifies the logical steps.
The specification is the contract. A proof that a controller never commands a forbidden speed is useful, but it does not prove the speed sensor is accurate or that the forbidden-speed rule covers every hazard. Specifying the right behavior remains human engineering work.
A controller for a laboratory instrument includes a checked guarantee that certain incompatible actuators cannot be enabled at the same time, under a stated model of its hardware.
Guarantees cover the stated specification and assumptions—not every interaction with an unpredictable world.
Background, not proof this scenario is available.
Solving math · Product
Already emerging
Libraries that run particular scientific and AI calculations faster without replacing the hardware.
Much of modern computing repeatedly performs a small set of operations, such as multiplying matrices. A better algorithm can reduce the work required for one of these operations without changing the chip. AI can search for mathematical shortcuts, then benchmark promising candidates on the hardware where they will run.
The useful product would be a tested library update, compiler improvement, or specialized kernel—not a new physical processor. A faster kernel only accelerates the part of an application that uses it: data movement, networking, and other calculations may still dominate.
An engineering team runs the same simulation, at the same required accuracy, on its existing servers. A new numerical library shortens some calculations; the team measures the entire job before deciding whether the upgrade is worthwhile.
Illustrative calculation, not a measured benchmark or forecast. A faster part does not speed up the whole task equally.
Overall speedup: 1.33×. The same work takes 75% of the original time (25% less).
Both bars use the same scale: 100% is the original total execution time. The model assumes no extra overhead and the same amount of work.
Total speedup = 1 / ((1 − p) + p / s)
p is the share of original time accelerated; s is its speedup. At p = 50% and s = 2, the task is 1.33× as fast—not 2×.
Gains on particular operations must translate into reliable, end-to-end improvements; this is not a universal speedup.
Background, not proof this scenario is available.
Modeling cells · Cures & treatments
Breakthrough-dependent
Better control of cell identity could help repair damaged cartilage or other tissues in selected conditions.
Regeneration requires more than producing the right cell type. Cells must form an appropriate structure, interact with their surroundings, and remain functional over time. Modeling could help researchers choose signals, scaffolds, and conditions that support that process.
Some tissues are especially difficult to restore because their architecture, mechanical properties, or blood supply are complex. A laboratory culture that resembles healthy tissue is an encouraging intermediate result, not proof of a durable repair in a person.
Researchers combine cell-state predictions with scaffold experiments to seek a cartilage repair that integrates with surrounding tissue and withstands repeated loading.
Organ-level structure, blood supply, immune compatibility, durability, and cancer risk are not solved by cell-state prediction.
Background, not proof this scenario is available.
Modeling cells · Cures & treatments
Breakthrough-dependent
Some cancers become more curable through combinations that attack a tumor's vulnerabilities while sparing healthy tissue.
A tumor is not a uniform population of cells. Different clones can respond differently to treatment, and resistant populations can become dominant. Predictive models might help identify vulnerabilities shared across those populations or combinations that reduce the chance of escape.
A promising combination must also spare enough healthy tissue to be tolerable. Understanding a cancer cell in isolation is insufficient: immune interactions, tissue context, and the distribution of drugs through the body affect whether a treatment succeeds.
For a particular cancer subtype, researchers investigate whether blocking two complementary pathways suppresses both the main tumor population and a resistant subpopulation. The model proposes the hypothesis; experiments and trials determine its value.
Tumor evolution, immune effects, drug delivery, toxicity, and clinical trials remain essential. This is not a universal cancer cure.
Background, not proof this scenario is available.
Modeling cells · Cures & treatments
Breakthrough-dependent
Small patient populations gain more candidate therapies, including for some disorders caused by a single gene.
For some rare disorders, a known genetic change offers a starting point, but the steps from that change to illness are poorly understood. Models could help identify which cellular functions are disrupted and whether restoring a function might help.
Possible interventions include existing drugs, new molecules, or genetic approaches, depending on the disorder. None follows automatically from identifying the mutation. Reaching the affected cells and intervening at the right stage of disease can be the hardest problems.
A team studying a rare enzyme deficiency uses patient-derived cells and computational models to investigate whether an existing compound can restore a measurable function. A reproducible laboratory result supports further development, not immediate prescribing.
Delivery to the right tissue, off-target effects, long-term safety, and trial feasibility still constrain progress.
Background, not proof this scenario is available.
Modeling cells · Insight
Plausible next step
Biologists distinguish changes that drive disease from changes that merely accompany it.
Large biological datasets reveal many differences between healthy and diseased samples. Some differences drive disease; others are responses to it, or reflections of unrelated differences between the people sampled.
AI can help prioritize which relationships to investigate and which interventions would distinguish competing explanations. A stronger causal claim comes from changing the suspected driver and observing the predicted consequences under appropriate controls.
If a gene is unusually active in diseased cells, researchers perturb it and measure whether the disease-relevant function changes. They also test an alternative explanation, such as inflammation causing both observations.
Causal claims need controlled interventions and replication; observational correlations alone are insufficient.
Background, not proof this scenario is available.
Modeling cells · Product
Plausible next step
Researchers screen candidate interventions computationally before spending scarce laboratory time on the most promising ones.
A useful virtual-cell model would predict selected responses to a proposed intervention: changes in gene activity, cell survival, or a disease-relevant function. It need not simulate every molecule in a cell to help researchers choose a more informative next experiment.
Predictions are most useful when they distinguish promising candidates from uninformative ones and expose uncertainty. The process remains a loop: predict, test in real biological systems, compare the result, and improve the model.
Researchers studying a disease pathway use a model to rank candidate compounds and doses. They test a carefully chosen subset, including controls and candidates the model is uncertain about, before deciding what to study next.
Estimate a specific cellular response and its uncertainty.
Prioritize informative interventions and controls.
Measure what actually happens in a biological experiment.
Use agreements and failures to revise the model.
Feed the results back into the next round of predictions and experiments.
Reliable prediction for unfamiliar compounds, cell types, and patient populations is still difficult.
Background, not proof this scenario is available.
Generating videos · Product
Already emerging
Small teams create concept trailers, visual effects shots, and ambitious short-form work with fewer production resources.
Generative video can make it practical to try shots that would otherwise require a location, a set build, or specialized effects. The creative leverage is often greatest early on: testing the visual language of a scene before a production commits to it.
A film still needs writing, performances, editorial rhythm, and continuity. A workable production pipeline must also make individual shots controllable enough to revise a character, action, or camera choice without losing the elements already approved.
A small team develops an ambitious science-fiction short. It uses generated concept shots to refine the look, then combines selected generated effects with performed footage, editing, and sound.
Long-form continuity, performances, editable assets, and production rights still need deliberate human work.
Background, not proof this scenario is available.
Generating videos · Insight
Plausible next step
Museums and classrooms explore reconstructed streets, buildings, and daily life through clearly labeled visualizations.
Historical reconstructions can combine photographs, maps, surviving objects, and scholarship into a navigable or cinematic explanation. Their educational value grows when visitors can see which parts are documented and which are interpretations.
AI should help express the evidence, not quietly fill every gap with plausible-looking detail. Showing two credible reconstructions of an uncertain feature may be more informative than presenting one confident image.
A museum visualizes a street at a particular date. Visitors can compare a reconstruction with source photographs and open notes explaining where signage, interiors, or daily activity are inferred.
Reconstructions must distinguish documented facts from interpretation. Generated footage is not historical evidence.
Background, not proof this scenario is available.
Generating videos · Product
Plausible next step
A teacher requests an age-appropriate animation explaining a pump, a cell, or a historical process.
A well-chosen animation can expose relationships that a static paragraph hides: how a valve moves, how a force changes direction, or how parts of a biological process interact. Generation could make such explanations easier to adapt to a learner's level.
The challenge is explanatory accuracy, not cinematic polish. An attractive animation that shows the wrong causal sequence can teach a misconception more effectively than an unclear diagram.
A teacher produces a slow, labeled animation of a pump cycle, then a second version that pauses before each valve movement and asks the learner what will happen next.
Subject experts must check accuracy; visual plausibility is not evidence that the depicted mechanism is correct.
Background, not proof this scenario is available.
Generating videos · Product
Plausible next step
A technician or homeowner sees an explanation tailored to the documented equipment and repair step.
Many practical instructions fail because the reader cannot connect the words to the object in front of them. A generated visual guide could highlight the correct part, show the sequence, and explain what successful completion should look like.
The guide must be grounded in the actual model and a validated procedure. If the system is uncertain which machine it is looking at, asking the user to identify it is better than generating a convincing but incorrect demonstration.
A user identifies the exact model of an air purifier. A guide based on its manual shows how to replace a filter, which side faces inward, and how to confirm it is seated correctly.
Safety-critical steps must be grounded in the actual equipment manual; invented details can make a demonstration unsafe.
Background, not proof this scenario is available.
Generating videos · Product
Already emerging
Creators localize selected scenes and audiovisual material for audiences they could not previously reach.
Localization is more than translating words. A line may need different timing, tone, or cultural context to work for another audience. AI-assisted dubbing and audiovisual editing can reduce some of the mechanical work while people guide the adaptation.
Good localization preserves what the creator means rather than copying every surface detail. Review by fluent speakers remains important, especially for humor, idioms, emotionally sensitive scenes, or technical explanations.
An independent documentary is adapted for a new language audience with translated narration, reviewed subtitles, and revised on-screen graphics. The filmmakers retain approval over the final meaning and performance.
Translation quality, consent for likenesses and voices, and cultural nuance still need review.
Background, not proof this scenario is available.
Making games · Product
Already emerging
Individuals and small teams prototype games that once required several specialist disciplines.
AI-assisted tools can shorten the distance between a game idea and something playable by helping with scripts, placeholder art, dialogue, and routine debugging. A creator can test whether a mechanic is enjoyable before investing heavily in production.
The advantage is rapid iteration, not guaranteed quality. Game systems interact in surprising ways, and generated code or assets still need integration, review, and testing on the devices players will actually use.
A first-time creator builds a small puzzle game, watches a few people play, and revises the confusing rules. AI helps implement those revisions; the creator decides what makes the game worth playing.
Production-quality games still require debugging, design, art direction, and integration.
Background, not proof this scenario is available.
Making games · Product
Plausible next step
A child practices fractions, a language, or scientific reasoning through a game that changes with their progress.
An adaptive game could vary the next challenge based on what a learner appears to understand. A good system would distinguish a conceptual mistake from a slip and offer an explanation or easier intermediate step when appropriate.
Adaptation should serve a learning goal that educators can inspect. Making the game more absorbing is not the same as making it more educational, and learners should not be trapped in an endless cycle of easier rewards.
A fractions game notices that a learner confuses the size of a piece with the number of pieces. It introduces a visual comparison, then checks the idea again in a different context.
Learning gains must be measured; engagement alone does not establish educational value.
Background, not proof this scenario is available.
Making games · Product
Plausible next step
Technicians and emergency teams rehearse uncommon scenarios before encountering them in real life.
Training simulations let people rehearse rare or difficult situations without exposing a real system to their mistakes. AI could make those situations more varied and the simulated people more responsive.
The training value depends on the decisions and consequences being correct. A photorealistic emergency scene can be a poor simulator if it rewards behavior that would fail in the real setting.
A maintenance team rehearses diagnosing a fault while a simulated colleague supplies incomplete information. The exercise rewards correct checks and communication, not merely reaching an answer quickly.
The simulator's behavior and scoring must match real procedures; realistic images do not guarantee realistic physics.
Background, not proof this scenario is available.
Making games · Product
Plausible next step
Players get better-supported control alternatives, adjustable pacing, and explanations that suit their needs.
Accessibility needs vary widely. Some players need alternative controls; others need clearer audio cues, larger text, less time pressure, or an explanation of a confusing objective. AI could help developers offer and maintain more of these adaptations.
Players should remain in control of the changes. A system that unpredictably takes over or silently changes the rules can undermine accessibility, trust, and the satisfaction of playing.
A player uses switch controls and chooses slower timing windows. The game offers concise explanations when requested while preserving its core puzzles and allowing the player to adjust the assistance.
Changes need testing with disabled players and must remain predictable and controllable.
Background, not proof this scenario is available.
Making games · Insight
Plausible next step
Researchers explore how people coordinate under different rules and resource constraints.
Interactive scenarios can help researchers compare how different rules affect coordination: what information people receive, how they share resources, or how they resolve competing goals. AI can reduce the cost of constructing and varying these environments.
A simulated society is not a substitute for evidence about actual people. Results depend on participants, incentives, and assumptions; AI agents may behave very differently from humans even when their dialogue sounds natural.
A research team compares two ways of sharing a limited resource in an experimental game, measures participants' decisions, and checks whether the result survives changes to the rules and participant pool.
Behavior in a game does not automatically generalize to society; findings need independent evidence.
Background, not proof this scenario is available.
Designing materials · Product
Breakthrough-dependent
If suitable materials exist, low-loss components could transform some magnets, electronics, and power equipment.
A superconductor has distinctive electrical and magnetic properties below its operating limits. A practical material that worked near room temperature and ordinary pressure could remove some cooling requirements and enable new designs for selected devices.
The central uncertainty is whether a suitable material exists and can be made reliably. Even a convincing superconductivity result does not establish useful wire: critical current, magnetic-field tolerance, mechanical properties, and manufacturing all matter.
If a candidate material is discovered, independent laboratories reproduce its superconducting behavior. Engineers then investigate whether it can carry useful currents in a workable component under relevant conditions.
AI cannot guarantee a practical room-temperature, ordinary-pressure superconductor exists; synthesis and useful current capacity are additional hurdles.
Background, not proof this scenario is available.
Designing materials · Product
Plausible next step
Vehicles and stationary storage gain safer, longer-lived, or less resource-intensive battery chemistries.
A battery is a system of interacting components. A promising electrode can fail when paired with a particular electrolyte, or work well in a laboratory cell but degrade during fast charging. AI could help prioritize combinations and understand failure patterns.
There is rarely a single best chemistry. A stationary storage system, a car, and a wearable device place different weights on weight, volume, lifetime, safety, materials availability, and cost.
A team designing stationary storage searches for a chemistry with long cycle life and abundant inputs, even if it stores less energy per kilogram than a premium vehicle battery.
| Desired improvement | What can pull the other way |
|---|---|
| More energy per kilogram | Must also satisfy safety, lifetime, and practical packaging requirements. |
| Faster charging | Can increase heat or degradation unless chemistry and thermal management support it. |
| Longer operating life | May require different materials or a more conservative operating window. |
| Abundant, inexpensive inputs | Still need suitable performance and a scalable manufacturing process. |
Energy density, cycle life, safety, charging rate, cost, and manufacturability must work together.
Background, not proof this scenario is available.
Designing materials · Product
Plausible next step
Useful chemicals are manufactured with less energy or fewer unwanted byproducts.
A catalyst can change the rate or selectivity of a chemical reaction without being consumed in the overall reaction. Discovering a useful one requires balancing activity with stability, availability, and behavior under actual operating conditions.
AI can guide the search by comparing candidate structures, reaction pathways, and experimental results. A catalyst that looks excellent with clean laboratory inputs may lose performance when exposed to impurities or repeated use.
Researchers investigate catalysts for making an important chemical at less demanding conditions, measuring not only reaction rate but also unwanted products and performance over time.
Activity in a model or laboratory must survive real feedstocks, long operating periods, and industrial economics.
Background, not proof this scenario is available.
Designing materials · Product
Plausible next step
Water systems remove selected contaminants or salts with improved energy use and durability.
A membrane separates substances by allowing some to pass more readily than others. Researchers must balance throughput, rejection of the target contaminant, operating pressure, and resistance to fouling.
AI could help identify promising structures and connect laboratory measurements to operating conditions. A good result with a clean test solution is only a first step because real water contains mixtures that interact with the membrane.
A water-treatment team tests candidate membranes for removing a particular contaminant, then runs them on representative feed water to measure rejection, flow, fouling, and cleaning requirements.
Fouling, maintenance, disposal, pressure requirements, and system costs still matter.
Background, not proof this scenario is available.
Designing materials · Insight
Plausible next step
Scientists learn which structural features produce valuable properties, allowing discoveries to transfer between material families.
A predictive model may identify relationships between a material's structure and a useful property. Scientists can investigate whether those relationships reflect mechanisms that remain valid beyond the examples used to train the model.
Failed experiments are informative too: they help define where a rule stops working. The ambition is a design principle that guides a new search, rather than a model that only recognizes close relatives of known materials.
A team proposes that a particular structural feature improves ion transport. It deliberately tests materials that separate that feature from other correlated properties, refining the explanation when predictions fail.
A predictive pattern must survive new compositions and conditions before it becomes a dependable design rule.
Background, not proof this scenario is available.
Robotics · Product
Breakthrough-dependent
Assistive robots bring objects, help with selected routines, and support people who need physical assistance.
An assistive robot could support everyday independence through selected tasks such as retrieving an object or positioning a lightweight item within reach. Its usefulness depends on whether it adapts to the person's preferences and makes assistance easy to request.
This is not a replacement for human care. Work involving close physical contact, lifting people, or consequential health decisions requires a different level of validation from bringing a book or carrying a bag.
A person requests a familiar object through an accessible interface. The robot retrieves it, confirms where to place it, and offers a simple way to stop or redirect the task.
Dependability, privacy, safe contact, human care, and access matter more than impressive demonstrations.
Background, not proof this scenario is available.
Robotics · Product
Plausible next step
Home robots handle a useful subset of repetitive tidying, carrying, and cleaning tasks.
The most useful household robot may begin with a few reliable jobs rather than an unlimited list of capabilities. Carrying familiar objects between known places or tidying a defined area can be valuable if it works repeatedly without constant supervision.
Real homes contain changing lighting, clutter, pets, and fragile objects. A robot must recognize when a task exceeds its ability and hand control back gracefully instead of turning uncertainty into damage.
A robot clears a designated table of familiar, non-fragile items and places them in assigned locations. It asks for help with an unknown object rather than guessing how to handle it.
Unfamiliar homes, fragile objects, stairs, pets, and failure recovery make broad reliability hard.
Background, not proof this scenario is available.
Robotics · Product
Plausible next step
Robots take on more work in contaminated, unstable, or hard-to-reach environments.
Robots can collect information in places where access is difficult or hazardous, such as damaged structures or contaminated facilities. Better perception and navigation may expand the range of situations they can handle.
The distinction between detecting a possible problem and making a reliable diagnosis matters. A visual anomaly may need confirmation with another sensor or inspection method before someone decides how to repair it.
An inspection robot surveys a hard-to-reach structure, flags a suspicious area, and records its location and sensor readings so an engineer can decide what follow-up is needed.
Communication, power, rugged hardware, and recovery plans still determine deployment.
Background, not proof this scenario is available.
Robotics · Product
Already emerging
Automated instruments and robots execute repeatable experiments, while researchers guide the questions.
Automated instruments can repeat bounded procedures, while AI helps decide which experiment would be most informative next. Connecting the two can create a closed-loop laboratory that refines a hypothesis or searches a design space.
The system needs scientific discipline, not just speed. Controls, calibration, traceable samples, and reproducible measurements matter as much as an algorithm's ability to suggest the next candidate.
A materials lab repeatedly prepares and measures candidate formulations. The next batch is chosen using previous results and uncertainty, while control samples reveal whether equipment drift is distorting the measurements.
Define a target property, constraints, and experimental controls.
Robots and instruments execute a validated procedure.
Record results, uncertainty, and quality-control failures.
Use the evidence to refine or challenge the current hypothesis.
Feed the results back into the next round of predictions and experiments.
Automation works best in bounded setups; calibration, reproducibility, and human scientific oversight remain necessary.
Background, not proof this scenario is available.
Robotics · Product
Plausible next step
Robots target weeds and inspect crops more selectively, potentially reducing waste and chemical use.
Agricultural robotics can turn a broad intervention into a local one: treating a weed rather than an entire area, or inspecting a plant rather than relying only on a field average. AI helps identify what the robot is seeing and select an appropriate action.
Farms are highly variable workplaces. A system that works in one crop, season, or soil condition may need substantial adaptation elsewhere, and a slow robot may be impractical even if its individual decisions are accurate.
A robot distinguishes crop plants from weeds in a defined growing system and applies a targeted mechanical intervention, with field measurements checking crop damage and actual weed control.
Weather, terrain, crop variety, throughput, and economics vary widely between farms.
Background, not proof this scenario is available.