Artificial Intelligence a Decade from Now: Three Possible Paths

About nine years separate us from 2035. This is not a space for contemplating destiny from afar, but a cramped room in which great decisions are made while we are busy with details. The most dangerous thing about the rhetoric of “technological determinism” is that it excuses us from responsibility for the present; the future is not a prophecy to be recited but choices to be made, and the unseen belongs to God alone. Our aim is not to add one more prophecy to a market crowded with them, but to dissect the logic behind three paths that researchers dispute.

The three paths

1. Continuous acceleration: models keep expanding until they approach a general intelligence matching humans in most cognitive tasks.
2. Plateau: capabilities hit diminishing returns and steep costs in energy and data, so progress settles at a useful ceiling short of the dazzling promises.
3. Societal stumbling: the limit is not technical but social and political; the technology falters at the boundaries of trust, governance and the distribution of power.

The One Hundred Year Study on Artificial Intelligence (AI100, 2021) notes that media coverage often exaggerates AI’s potential, for better and for worse, and that some optimistic forecasts, such as fully autonomous driving, have not come true.

Why are there several paths?

Behind the dispute lie the so-called “scaling laws”: a model’s loss improved predictably as its size, data and compute grew (Kaplan et al., 2020). But this is a limited empirical observation, not an eternal law, and no one guarantees it will continue. The field has also known “AI winters”: waves of enthusiasm followed by disappointment whenever a promise stumbled at the limit of the possible.

Three dimensions are mixed together in public debate:

• Technical: is there a known engineering route to general intelligence? The honest answer: no one knows.
• Cognitive and philosophical: has the machine come to understand, or does it merely imitate? (See: “Three Myths About Artificial Intelligence”.)
• Social and political: computing power is concentrated in a limited number of companies and states, and whoever owns it holds the keys of influence.

The speed of innovation is not the speed of absorption

Between an innovation’s birth in the lab and its pervasiveness in life lies a time gap, filled by legislation, education, the reorganisation of work and the building of trust. The waves of electricity (David, 1990) and the computer showed that the great returns arrive after a decade or more, when institutions redesign themselves around the tool. So what we see in 2035, God willing, will most likely reflect how well we localised the technology more than its maximum capability.

What the “fate of humanity” rhetoric leaves out

• Most likely the three paths will materialise together: acceleration in sectors with clean data and abundant capital, such as pharmaceuticals, finance and programming; a plateau in tasks that demand an understanding of the physical world or complex ethical judgement; and stumbling in societies that bring in the technology faster than they build its governance.
• Preoccupation with the superintelligence scenario diverts attention from real, immediate risks: bias, concentration of wealth, manipulation of opinion and the widening digital divide. Some critics hold that inflating existential risk distracts from harms already present.
• Conversely, rejecting concern outright is naivety of another kind.

Estimates of the occupations threatened by automation vary widely with method: Frey and Osborne estimated that about 47% of US jobs were exposed to automation, whereas an OECD study using a task-based approach concluded about 9%. The wiser course is to ask: which skills will be rewritten, and how do we prepare people for what the machine does not do well?

A more useful question: what structure makes us winners on any path?

A nation that builds its human and institutional capacity firmly benefits from acceleration if it happens, withstands a plateau, and absorbs stumbling. Four pillars:

1. Immunity that holds on every path: deep education, original research, national computing infrastructure and entrepreneurship.
2. Sound governance with effective human oversight, as the OECD principles state (see: “Responsible Artificial Intelligence”).
3. Human capital: the highest-return institutions are those that redesign work so the human complements the machine, not those that replace their staff.
4. A critical sense in the young that tells the genuine from the fake; among the gravest threats to societies is a person who has lost the ability to detect lies.

Saudi Vision 2030 offers a model worth reflecting on in building national capability from the ground up, through education and research, rather than importing it ready-made.

So which path are we taking?

The question, put this way, is wrong from the start. Most likely 2035 will not be the result of one scenario triumphing, but the sum of millions of small decisions we make today in our classrooms, laboratories, legislatures and boardrooms. And since the decision is in our hands, we ask God for success and take the means; the future, God willing, is not awaited but pursued.