When Your Date Is an AI Clone

Now, we can clone ourselves on a dating app.

As a NY Times reporter described, he began by identifying the smaller start-ups that offered cloning. Next, through several “training” conversations, he familiarized a bot with his personality and the sound of his voice. Then, the clone was ready to look for love–perhaps with another clone. Sadly, when his clone spoke with Rachel on their first date, he sounded too much like a customer service chatbot.

Knowing how much and how fast artificial intelligence has spread, a European think tank asked if the investment is generating a sufficient return.

Artificial Intelligence Spending

In a recent working paper, Europe’s Bruegel think tank expressed concern with artificial intelligence. Focusing on productivity, they expect a delay between implementing AI and its productivity payoff. The problem is the tension between gargantuan spending and minimal revenue.

We can start with petaflops.

In 2017, spending for less than 10,000 petaflops was close to $1,000. Now though, taking the leap to more than 100 million petaflops, the cost exceeds a whopping $200 million:

The costs include training and the hardware for the data centers that would compose an AI infrastructure. Explaining the cost explosion, the Bruegel paper says that AI’s cognitive improvement needs an accompanying rise in complementary inputs. Below, you can see specific spending requisites:

The Bruegel researchers tell us that the increase in AI will be fueled by individuals, firms, and countries.

Altruism will motivate people to adopt AI because of the “good” it can do for areas that include the environment, health, and transport. Meanwhile firms are nudged by the “chicken game factor” or we could call it FOMO. Then, with countries, the Bruegel researchers cite China worries that relate to the military and security innovation.

Our Bottom Line: AI Productivity

Bruegel suggests that all of the cost estimates can only be validated by AI generated productivity. And that is where they have questions.

As always, we can define productivity by comparing input to output. Oversimplifying, we can just say that we are more productive when less land, labor, and capital produce relatively more goods and services. For AI, we certainly will become more productive when machines replace humans. Soon-to-be Nobel economics laureate Daron Acemoglu estimates a .5% AI inspired productivity increase during the next decade. Disagreeing, the Goldman Sachs productivity number was 10%. Whichever is correct, we have awhile to wait.

Returning to where we began, we can also ask if a dating clone makes us more productive?

My sources and more: Thanks to Slate Money for alerting me to the NY Times AI dating article. From there, having read this Bruegel (think tank) paper on AI and productivity, I had the perfect complements.