My top 3 lessons* on artificial intelligence from Erik Brynjolfsson
There is no cure for curiosity. I had to admit my failure to understand all the nuts and bolts of artificial intelligence (more on that here). However, I am even more curious about the effects AI is going to have on the economy and our lives in general.
I hoped that Erik Brynjolfsson who is the director of Stanford Digital Economy Lab would be the person who will talk more about the effects rather than the inner workings of the AI. And I was not disappointed.
Here are my 3 takeaways from watching Erik’s presentation at the Wharton AI workshop on YouTube:
Lesson #1
Traditionally computers have required inputs (x) and instructions (f(x)) to produce outputs (y). Machine learning is different – it requires inputs (x) and output (y) to learn instructions (f(x)). After the instructions have been learned they could be applied to new inputs to get new outputs. To do these inferences you need huge amounts of data, enormous computing power and improved algorithms. Technically, you can call these the “prediction machines” because they are using inputs to predict outputs.
Lesson #2
Recent studies conducted on the use of AI at call centres suggest that the main beneficiaries of AI assistance are lower-skilled employees and new hires (employees were free to choose whether to use the responses proposed by AI in real-time during the calls or not). Having studied the best practices of top performers AI made those skills available to all employees. The results of the studies demonstrate that the AI support allows employees to catch up faster resulting in a very rapid performance improvement.
It is interesting that contrary to the initial hypothesis AI didn’t become a crutch. An AI outage that happened during the experiment demonstrated that people had learned from the AI. Even the language of lower-skilled employees had changed in the process and they started to sound more like their experienced colleagues.
Lesson #3
Digital progress will make the economic pie bigger. However increased productivity is not an absolute benefit, it can destroy profits.
With labour hours approaching zero, productivity can theoretically increase to infinity. But that doesn’t solve the issue of labour income which will be missing. The situation, if not properly managed, could resemble the 1930s when new farming methods and machines radically improved productivity but the demand was simply not there. As a result, the farmers, while highly productive, were earning less money than before.
And, as a final note, comes a warning from Erik. In 2022, Metaculus, an online prediction aggregation engine, suggested that Generative AI will be tested and formally announced and launched in 2057. In 2023 the forecast brought this date forward to the year 2040. Now the latest prediction earlier this year claims that all this is going to happen already in 2032. This might be the time to start taking it seriously.
For those willing to try and pick top lessons of their own here is the link.
* from anything that you are reading, watching or hearing you can realistically expect to remember a limited number of things only. My solution is to pick just 3 items or ideas from any material. This number is non-negotiable. Even the most extraordinary experience has to be compressed into 3 things to remember. This approach has worked for me so far.
This note was originally published on Medium.com on 13 October 2024
