Artificial intelligence has gone through many cycles of excitement and disappointment. Today we are in an upswing, thanks to the development of large language models (LLMs) based on the Transformer architecture. This architecture is particularly well suited for GPUs, making it possible to use immense amounts of data and computing power to train models with billions of parameters.The most notable consequence is the creation of a new user interface for computers: human language.
Just as the graphical user interface made the personal computer accessible to millions of users in the 1980s, new natural language interfaces have made AI accessible to hundreds of millions of users worldwide in the past year.
Despite this apparent accessibility, the "democratization" promised by SaaS solutions remains imperfect and partial, creating new forms of inequality.
AI still requires specific skills:
- AI literacy and understanding the limitations of systems
- Ability to critically evaluate outputs
- Business process integration skills
John McCarthy coined the term AI in the 1950s, but he himself complained, "As soon as it works, no one calls it AI anymore." This phenomenon, known as the "AI effect," continues to influence us today.
The history of AI is littered with successes that, once they become sufficiently reliable, are no longer considered "smart" enough to merit the aspirational appellation.
Examples of technologies that were once considered cutting-edge AI and are now taken for granted:
- The computer vision that is now built into every smartphone
- Voice recognition, now simply "dictation"
- Language translation and sentiment analysisRecommendation systems (Netflix, Amazon) and route optimization (Google Maps)
This is part of a larger phenomenon that we can call the "frontier paradox."
As we attribute to humans the frontier beyond our technological mastery, this frontier will always be ill-defined. Intelligence is not something we can capture, but an ever-approaching horizon that we turn into useful tools.
The spread of generative AI has drastically reduced the cost of producing and transmitting information, with paradoxical effects with respect to the goals of civic participation.
The combination of generative AI and social media has created:
- Cognitive overload and amplification of pre-existing biases
- Increased social polarization
- Ease of manipulation of public opinion
- Proliferation of fake content
Simplified interfaces hide the workings of AI:Poor understanding of automated decision-making processesDifficulty identifying algorithmic biases
Limited customization of underlying modelsThe importance of human-led automated intelligenceAI can only take us 90% of the way there.
Machines excel at analyzing large volumes of data, but struggle with edge cases. One can train algorithms to handle more exceptions, but beyond a certain point the resources required outweigh the benefits. Humans are precise thinkers who apply principles to edge cases, while machines are approximators who make decisions based on prior
As described by Gartner in technology hype cycles, wild enthusiasm is invariably followed by disappointment-the "valley of disillusionment."
Founders benefit in the short term from eye-catching marketing, but at a cost.Alan Kay, computer science pioneer and Turing Prize winner, said, "Technology is technology only for those born before it was invented." Machine Learning professionals are scientists and engineers, yet their efforts always appear like magic-until one day they are not.
Homogenization and loss of competitive advantageWidespread adoption of the same pre-built SaaS solutions results in:Convergence to similar business processesDifferentiation difficulties through AIInnovation limited by platform capabilitiesData persistence and its risks
With the accessibility of generative AI platforms:Data persist over time in digital infrastructureData points can be reused in different contexts
A dangerous cycle is created when future generations of AI are trained on synthetic content.
The AI market is dividing into:
- Commodity AI: standardized solutions available to many
- Proprietary advanced AI: cutting-edge capabilities developed by a few large organizations
Part of the problem lies in the very definition of "Artificial Intelligence."
If we break down the term recursively we find that each branch of the definition refers to "humans" or "people." By definition, then, we think of AI as imitative of humans, but as soon as a capability firmly enters the realm of machines, we lose the human reference point and stop considering it AI.
It is more useful to focus on specific technologies that can be put to work, such as transformers for language models or diffusion for image generation. This makes our ability to evaluate an enterprise much more explicit, tangible, and real.
The frontier paradox means that AI is accelerating so rapidly that soon it will simply be technology, and a new frontier will become AI. Becoming "technology" should be seen as recognition for an idea that was previously at the cutting edge of the possible.This article was inspired in part by Sequoia Capital's reflections on the AI paradox.
For more information: https://www.sequoiacap.com/article/ai-paradox-perspective/
The real promise of accessible AI is not simply to make technology available, but to create an ecosystem in which innovation, control and benefits are authentically distributed.
We need to recognize the tension between access to information and the risks of overload and manipulation.
Only by maintaining a strong human element in artificial intelligence and adopting a more precise language can we realize its potential as a force for truly distributed inclusion and innovation.
.svg)
.svg)
.svg)
.jpeg)
.jpeg)