Artificial intelligence (AI) requires vast infusions of capital to flourish. This means it needs investors. But there’s a conundrum: in trying to gauge a return on an AI investment it’s hard to place a valuation on the technology. That is because AI offers the most mind-bending ways to accelerate productivity while simultaneously morphing in form, substance and utility.
AI is shape-shifting by the second (celluloid film technology, by contrast, barely changed for a century). How does one consider such an investment? Is it time to buy? Is the promise outstripping reality? Is it already too late? Sometimes, the best way to understand a conundrum is via a thought experiment. Let’s think about the value of an AI-powered friendship network as a way to understand the value of ever-expanding AI.
In an April interview with podcaster Dwarkesh Patel, Meta's Mark Zuckerberg cited a statistic that shows the average American has fewer than three friends. Zuckerberg reckoned the number should be closer to 15. A classic economic problem: supply is insufficient to meet demand. Naturally, Meta has the answer — it will deliver beautifully crafted AI companions. These will be digital beings, algorithmically forged to complement our habits, desires and interests.
On the face of it, for the individual, a large and expanding friendship network must have a richer valuation. Having many friends is surely better than having fewer. Once I had moved past my furious mental tally confirming how far I fell short of the magical 15, I was alternately amused and horrified at Zuckerberg’s predictable conclusion.
He was almost certainly attuned to the 2023 declaration by US surgeon-general Vivek Murthy that the country was experiencing an “epidemic of loneliness and isolation”. Murthy wrote of the mortality effect of social disconnection as “similar to that caused by smoking up to 15 cigarettes a day”. In the paper there was a helpful infographic defining social connection as comprising three parts — structure, or the number and variety of relationships and frequency of interaction; function, or the degree to which relationships serve various needs; and quality, or the positive and negative aspects of relationships and interactions.
Let’s use our imagination and take a little meander into Meta’s product development process. Picture a whiteboard: 207-million lonely Americans with mortality risk equivalent to 15 cigarettes daily. The surgeon-general has handed Meta the medical emergency of a lifetime with a droolworthy price tag. The product team breaks down social connection into component parts.
- Structure? Solved. Infinite AI friends (not merely 15), available 24/7.
- Function? Solved. Customise your companion roster: the life coach at 6am; the socialite at lunch; and the empathetic listener at midnight.
- Quality? Solved. Your AI companions agree with you, validate you, reinforce you and never disturb you in a crisis.
This sounds terrific. Except it’s not. Breaking friendship into neat parts misses what makes friendship whole. The messy bits. The unpredictable, frustrating, infuriating bits that, when soothed, bind us even closer together.
As the recent sycophancy debacle at OpenAI showed, while scientifically the quality of relationships matter, manufacturing an endlessly positive experience is not equivalent to a friendship where there is meaning and connection. When ChatGPT agrees with every opinion you express, it is temporarily comforting until you feel the gulf between algorithmic responsiveness and genuine human pushback.
Real relationships involve ups and downs. There is pain in the fight and because of the pain, deep fulfilment in the resolution. The valuation question becomes, of course, a question of what the market is pricing.
Which brings us back to the original conundrum: figuring out a return on AI investment is tricky. Will there be a market for AI companions? Absolutely. Will they fill parts of the void for many people? 100%. Will they cure Dr Murthy’s epidemic of loneliness and isolation and, in aggregate, extend the lifespan of humanity?
No. Because mathematical models are simply humanity’s brutish attempt to distil our uncertain world into composable parts. That representation should not be mistaken for our humanity, which is rooted in physicality. The randomness that physicality brings can never be captured by a simple — even 405-billion parameter plus — equation.
So, in placing a valuation on AI there are at least two key considerations. Is AI’s promise of limitless efficiency the simple end game? Or should we think a lot harder about how to (mis)price AI — uncertain as we are about how effectively it will complement the inimitable, quintessential and singular messiness of being human?
• Bassier is COO at Ninety One.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.