Between AI Utopia and Dystopia, There Is Choice
The future of AI may be uncertain.
Our responsibility in the present is not.
Two futures, one effect
Most narratives we hear nowadays about AI tend to fall on two sides: utopia or dystopia.
On one side, AI will cure diseases, solve climate change, transform education, create abundance and perhaps even free us from much of today's work.
On the other, AI will take our jobs, destabilize democracies, enable devastating cyberattacks, escape human control and, in the most extreme scenarios, threaten humanity itself.
Two radically different futures. But they have something interesting in common: both focus our attention on what AI might do tomorrow.
And both fuel the race today.
The utopian narrative creates FOMO (Fear Of Missing Out) around opportunity. We need to invest because AI will transform everything and we cannot afford to be left behind.
The dystopian narrative creates another kind of FOMO. We need to keep developing because if we don't build it, somebody else, another company or another country, will.
Both narratives therefore benefit the race itself and the mega corporations behind it. A polarized debate between extraordinary promise and extraordinary danger is actually a blessing for the FOMO god.
Instead of constantly debating what increasingly powerful AI might eventually do, what are we choosing to do with the AI we already have?
If we believe the promises, why wait?
The utopian side contains many promises.
AI could accelerate drug discovery and perhaps cure diseases. It could enable personalized education, scientific breakthroughs, better climate modelling, new materials, enormous productivity gains, shorter working weeks and perhaps eventually some form of abundance.
These aren't necessarily fantasies. Some may happen.
The problem is that their future possibility is frequently used to legitimize present acceleration.
If these are genuinely our objectives, why are they mostly promises about what increasingly powerful AI will eventually do?
We already possess substantial computing capacity, increasingly capable models, enormous pools of capital and some of the world's best scientific and technical talent.
So why wait?
If AI will someday help solve climate change, what percentage of today's extraordinary AI investment is deliberately directed toward climate solutions?
If AI will someday create better lives for workers, what conditions do today's data workers, moderators, miners, semiconductor workers and other people throughout the AI value chain experience?
If AI will someday create abundance, who captures the productivity gains being generated today?
If AI will someday make healthcare radically better, what are we deliberately doing with today's models, compute, capital and talent to address existing diseases and unequal access to healthcare?
These questions do not invalidate the promises, but ask us to start taking them seriously now.
We already have enough capability to deliberately allocate resources to cancer research, neglected diseases, climate adaptation, energy efficiency, education or other explicitly defined public-interest objectives.
We don't need to wait for AGI to start.
Don't promise purpose after capability. Build capability and purpose in parallel.
If we believe the warnings, why wait?
The dystopian narrative reverses the emotional charge but retains the same future orientation.
Catastrophic cyberattacks. Autonomous weapons. Uncontrollable systems. Deception at scale. Loss of human control. And ultimately, superintelligence capable of causing human extinction.
Again, these possibilities should not simply be dismissed.
But exactly the same question applies.
If autonomous AI might someday escape human control, what are we learning from today's agentic failures and unintended behaviours?
If AI could someday become impossible to audit, why isn't traceability being engineered into systems now?
If AI could someday destabilize democracy, what are we doing about synthetic persuasion, misinformation and concentration of informational power today?
If AI could someday enable devastating cyberattacks, what are we doing about the misuse, fraud, phishing and impersonation already being enabled or amplified today?
If AI could someday become a systemic threat, why are safety, resilience and governance competing for resources with capability development rather than developing alongside it?
We don't need speculative rogue superintelligence to justify action.
We already have concrete problems: concentration of power, misinformation, cybercrime, opacity, environmental costs, labour displacement, dependency, automation bias and systems operating at speeds institutions struggle to govern.
The problems we can already observe give us enough reason to act, experiment, learn and build safeguards now.
Don't promise safety after capability. Build capability and responsibility in parallel.
Responsibility is also an allocation decision
The responsible position on AI is therefore neither to believe all the promises nor to dismiss all the warnings. It is to stop using the future as an excuse for avoiding choices in the present.
If AI is as promising as the optimists claim, we should start directing it toward those promises now.
If it is as dangerous as the pessimists claim, we should start engineering against those dangers now.
Either way, responsibility cannot wait for AGI.
And this is where responsibility becomes very concrete.
We can decide where investment money goes.
We can decide where GPU and computing capacity goes.
We can decide where researchers and human talent go.
We can decide which applications receive attention.
We can decide how much energy and infrastructure we are prepared to dedicate to different objectives.
We may not control everything about the development of AI. But resource allocation is one of the most powerful choices we do control.
So, if it were my choice, I would start with something deliberately simple:
30% of AI resources directed toward pursuing the beneficial promises of AI now: cancer research, neglected diseases, climate adaptation, energy efficiency, education and other explicitly defined public-interest objectives.
30% directed toward addressing the negative consequences and potential risks of AI now: concentration of power, misinformation, cybercrime, opacity, environmental costs, labour displacement, dependency, automation bias, safety and resilience.
40% directed toward continuing frontier research, better models and the technical future of AI, even if that eventually leads to AGI.
Is 30/30/40 the scientifically optimal allocation?
Of course not. I don't think anybody could seriously claim to know the optimal ratio. The point is the principle behind it.
Today, capability development often comes first, while purpose, consequences and safeguards are expected to catch up.
I would stop treating these things sequentially.
Capability, purpose, consequences and safeguards should develop in parallel.
Who should do that?
Very simple: all of us, the Responsible Humans.
Of course, we don't all control GPU allocation, AI investment or frontier research.
But we are not powerless either.
Companies make investment and procurement decisions. Investors decide where capital flows. Researchers decide what problems deserve their attention. Governments establish incentives and regulation. International institutions can create coordination mechanisms. Employees can question how AI is deployed inside their organizations.
And as citizens, we can act through our votes and representatives, support initiatives at national and international levels, participate in forums and public discussions, and sometimes simply refuse to accept that the direction of AI development is something happening beyond our influence.
The level of influence differs.
The responsibility to exercise the influence we do have does not.
That, ultimately, is what I mean by Responsible Humans.
Executive Reflection
We may not know whether artificial intelligence will eventually cure cancer, create abundance, destabilize civilization or develop capabilities beyond our control.
But uncertainty about tomorrow does not absolve us from responsibility today.
We know where data centres are being built. We can measure their energy and water consumption. We can examine where minerals come from. We can see the people labelling data and moderating content. We can test how agents behave. We can observe where power is concentrating.
And we can decide where capital, compute, energy and human talent are directed.
The alternative between AI utopia and AI dystopia is therefore not another prediction about the future.
It is choice.
If we genuinely believe in AI's promises, we should invest in realizing some of them now.
If we genuinely believe in AI's risks, we should invest in addressing some of them now.
And if we believe that further capability development remains important, we can continue doing that too.
These objectives do not have to wait for one another.
The future of AI may be uncertain.
Our responsibility in the present is not.
Igor Allinckx
AI & Humanity
October 2026
Back to Insights
Part of an ongoing exploration of governance, AI, and human judgment.