Navigating the Cone of AI Uncertainty
As walking through this cone has been one of my most popular talks this year, I thought I’d share it in an edition of the newsletter. Regular readers will recognise it’s an evolution of an earlier cone concept.
As we navigate the evolving AI landscape, one of the biggest challenges facing leaders is the sheer uncertainty about where we’re headed. The hype cycles, conflicting predictions, and exponential pace of change make strategic planning can feel like shooting in the dark.
This uncertainty is particularly acute when we consider the sustainability challenges facing generative AI, from environmental impact to economic viability.
This is where the “Cone of AI Uncertainty” becomes useful—a framework borrowed from futurology (specifically Magnus Lindkvist in my case) that maps the range of possible outcomes we might see in the next 2-3 years.
Understanding the Cone
The cone concept acknowledges a fundamental truth about predicting the future: the further ahead we look, the wider the range of possibilities becomes. Rather than trying to predict one “correct” future, the cone maps five distinct scenarios across a spectrum from dystopian to utopian outcomes. Bear in mind multiple futures might also come true in different contexts/industries/geographies.
The Five Scenarios: 2025-2028
- Struggle to Deliver ROI Against High Costs (Bottom) The hype bubble deflates
In this scenario, organisations find themselves unable to justify (what have become) substantial investments required for AI transformation:
- High implementation and operational costs (which increase as tech platforms seek to recoup their investments) consistently outweigh demonstrated benefits
- The hype bubble bursts as reality fails to match the ambitious promises - we don’t move on from language and content generation models - so it lacks the world view and understanding required to be consistently useful
- Many AI initiatives are quietly shelved or scaled back significantly
- Focus shifts to more modest, proven applications with clear business cases - we go back to narrower applications of AI like we did pre GenAI
Look at content from the likes of Denis O., Gary Marcus etc if you want to tune into future signals that align with aspects of this scenario.
- “Enshittification” Cost reduction trumps user experience
The “enshittification” scenario describes what happens when AI gets deployed primarily to cut costs regardless of suitability:
- AI is used to reduce cost and complexity (or “just do AI as everyone else is” / “it will increase our share price”) without considering whether it’s the right tool for the job
- This then results in degraded customer experiences as organisations prioritise savings over service quality
- Systems work “well enough” from a cost perspective but frustrate users and customers
- The focus shifts from “does this improve outcomes?” to “does this reduce our expenses?”
- What regulators chooses to do in response to this will be interesting to observe
- Augmentation / “Medium AI” (Middle) Steady, pragmatic progress
This represents measured but meaningful progress, aligning with discussions about the importance of architecture in AI implementations:
- Disruption occurs in specific model and solution-capable tasks rather than wholesale industry transformation
- Efficiency improvements make AI accessible and genuinely useful across more organisations
- AI successfully augments human capabilities without wholesale replacement
- Focus remains on practical applications with demonstrable value
- AI Significantly Disrupts Industries Selective transformation with managed trade-offs
In this scenario, AI delivers significant but uneven disruption across sectors:
- Major transformation occurs in specific industries/roles (finance, marketing, legal services) whilst others see limited impact
- Organisations successfully navigate the profit/people/planet trade-offs through careful planning and regulation
- Winners and losers emerge, but transition support and retraining programmes help manage social impact
- Environmental costs are acknowledged and actively managed through efficiency improvements, focus on materiality, and policy intervention
- Change happens at a pace that allows adaptation rather than shock
- AI Hype Becomes Reality (Top) Massive industry disruption - but at what cost?
The scenario where AI delivers on its transformative promises, but the techno-optimist vision proves problematic. As discussed in “There is more than one way to do GenAI”, this isn’t necessarily the only path forward:
- Massive industry disruption occurs as predicted, fundamentally reshaping the economy
- However, the environmental costs prove substantial - clean power can’t be sourced at the scale required
- Energy and resource consumption accelerates faster than sustainable alternatives can be deployed
- Benefits concentrate amongst those who control AI infrastructure, exacerbating inequality
- Society struggles with the pace of change and unintended consequences of rapid transformation - see sleepwalking challenges
Why This Matters
The cone isn’t about predicting which scenario will occur. It’s about driving a conversation that acknowledges the range of possibilities and preparing accordingly. Bearing in mind this will be happening all across the ecosystem - from our suppliers to clients and stakeholders. Beyond the technical and economic considerations, we must also be mindful of the broader societal impacts we may be sleepwalking into as AI becomes more pervasive.
By mapping these scenarios, we can make decisions that remain robust across multiple futures rather than betting everything on the current technology industry narrative (aka hype!) or any other single predicted outcome. The recent developments in sustainable approaches to AI architecture suggest there may be more nuanced paths forward than the current “bigger is better” narrative suggests.
The architecture and business choices we make today will influence which part of the cone becomes reality for our organisations. We need to get more conscious and conscientious about the future we are creating.
Do leave a comment or send me a DM if you’d like to discuss further or you are interested in having me present something like this at an event.
Originally published on LinkedIn. Comments and discussion live there.