"Any sufficiently advanced technology is indistinguishable from magic." — Arthur C. Clarke
"Technology is a word that describes something that doesn't work yet." — Douglas Adams
The promised land of AI
Artificial Intelligence has become a potential key to unlocking value creation. It is now often top of the agenda to unlock the $3.2 trillion traffic jam in Private Equity exits.
From my perspective, AI represents a generational change in technology, on par with the web, mobile, cloud, and social media revolutions.
Our natural instinct to view intelligence as a human characteristic leads us to evaluate these AI technologies through a human lens. And this is where the fun and weirdness starts.
When we look at AI, are we ready to ponder if "it" has a body, a mind, and a soul?
Does "it" have a body?
“Crazy evil genius” Elon Musk unleashed an army of robots at his latest promotional Tesla event launching Robotaxis.
The robots walked, talked and served drinks to the press. Everyone was amazed by those robots. The future is here! We will soon have a robot in every home!
However, it was later revealed that these robots were actually remote-controlled by event staff and the amazing interactions where robots have a chat with you and serve you a drink were an illusion.
This should not come as surprise considering Elon hosted an event 3 years ago where he introduced the futuristic robots and their capabilities by having a dancer dressed in a robot costume.
So it would be easy to be disappointed by the limited progress in humanoid robots. But then we saw, in the same week, how SpaceX's robotic arms were catching the returning Starship rocket booster at the launch pad. A truly amazing feat of engineering and software that will change space travel.
So sometimes we want our futuristic machines to be humans with a body, but we forget how amazing these machines already are.
Does "it" have a mind?
I have been running and advising data science teams for the last 10 years, and the pace of progress is truly phenomenal.
However, when it comes to assessing the reality vs the potential of the technology, there is a always a question to be asked: Can "it" truly reason or are we looking at probabilistic machines that are directionally correct but precisely wrong?
A recent study on mathematical reasoning in LLMs revealed some interesting findings:
LLMs remain sensitive to changes in proper names and numbers in problem statements. I.e. a math problem: “Mary had 5 coins and then spends 3, how many coins does Mary have?” gets a different (and wrong) answer if Helen is the one with the coins.
Including irrelevant information in math problems causes performance to drop across all models. I.e. introducing a comment about the air temperature that day might change the answer on how many coins Mary has.
As questions increase in difficulty, not only does performance drop, but variance also increases, making models less reliable. I.e. good students don’t do predictably better than average students when answering harder questions.
While workarounds and new tools like chain-of-thought reasoning will help address these issues, we need to adjust our current expectations about AI having a "mind" like humans do.
Therefore, it is currently challenging but not impossible to use LLM's top capabilities in situations where accuracy and consistency is critical. e.g. healthcare or regulated financial services.
This does not take away from the huge progress in repetitive predictable tasks that have large training datasets and require no or low reasoning such as customer service scripts and routine data crunching or code development.
Does "it" have a soul?
There is probably nothing as intrinsically human than music.
Music can transform how we feel, boosting our joy and healing our sorrows. The connection that we feel with our favourite singers and bands is unique.
So what happens when the AI starts singing?
This is happening in the US Billboard charts with a country song where the singer cannot speak in the real world due to a medical condition but AI recreates his voice from past records.
It also happened in Germany with a viral europop song that was AI generated and got into the Top 30 charts on Spotify.
This progress is unstoppable. New software like Suno (aka ChatGPT for Music) makes it so easy for anyone, irrespective of talent, to push a button and have AI create professional quality songs.
So, does "it" have a soul that connects with us? Will it win a Grammy?
Or is it all a trick, a mathematical probabilistic model that generates the music you want to hear the same way gmail autocompletes your email responses predicting your needs?
It's a kind of magic
As the AI technology develops exponentially, it is easy to fall into the trap of constantly comparing to a human body, mind or soul. We are wired to use a human lens when assessing something that approaches intelligence.
We might be better off thinking of AI as a magician's illusion that keeps improving until you can't figure out how the illusion is made.
It's the illusion of humanity, not humanity itself – for now.
I sincerely believe AI will be one of the most transformational technologies of our lifetime, I also believe we are not yet ready to fully absorb its implications for what it means to be human.
In the meantime, let’s enjoy the magic show,
Juan Lopez-Valcarcel
PS: This was a light-hearted view on AI. For an in-depth look at AI technology progress and its future, I highly recommend the always outstanding annual "State of AI report” by Nathan Benaich.