This blog post is about a talk given by Prof. Michael Jordan at SysML conference on the 15th of February.. He’s a professor of Statistics in the Department of Electrical Engineering and Computer Science and the Department of Statistics at the University of California, Berkeley. He’s extremely well-known, and has over 130,000 citations on google scholar. This is very much a follow on post from my previous blog post The Two Cultures of Data Analysis. All the quotes here are from him in the video and slightly paraphrased due to context.
He first mentioned that part of the reason he chose statistics back when he was choosing what to study > To me computer science was really exiting and all but it hadn’t really embraced the problems of inference, and to this day I still think there’s disconnect.
He then mentioned a scenario where he was then acused of being in Artificial Intelligence (AI).
No I’m not in AI, I’m in machine learning. Isn’t that just AI? No, it’s really much much broader than that…
We don’t have AI yet.
This is, in my opinion, an extremely accurate statement. This AI term seems to be thrown around in the press for even the most simple examples which results in some fake news.
We don’t have AI we just have systems that take input data and output data, they just mimic AI they aren’t actually intelligent. Guys, that’s not intelligence, we don’t have it yet, so for us to be using that buzzword we’re selling something to the public that we don’t have.
This is actually quite interesting as most the time when you see AI in an article, and they explain how great it is and all the possibilities, it’s actually just a neural network or some algorithm that has learned something from a specific dataset and has shown to do well. This is extremely cool, but when things change in real life and that data is longer from the same system as the current real-world data the algorithm is useless. Saying these things are AI is extremely inaccurate, for it to be intelligent it must be able to adapt.
He continuously comes back to this fact that we’re selling AI and calling it fake news. He gives a perspective on where the AI buzzword came from:
We’ll it was John McCarthy… John McCarthy arrived at MIT, and he says I want to work on intelligence in computing, and they said well isn’t that cybernetics, we already have x that does that, he couldn’t really convince people it was based on logic more than control theory, signal processing etc. So, he had to give it a new buzzword, so he called it Artificial Intelligence.
Types of “AI”
AI was not pursued with great success but Intelligence augmentation was pursued with great success.
He gives the perspective that what we’ve been doing isn’t AI, it’s been targetting the aspects of intelligence he terms “IA” and “II”.
What is real AI
He describes this as the “I, robot” problem, where you have one “machine” that can solve all the worlds problems and learn itself. This is the kind of AI in the films such I, Robot. They actually learn about their surrounding and seem to be able to genuinely react as humans would. We are extremely far from this.
IA, intelligence augmentation
He notes that intelligence augmentation is in essence a machine learning algorithm. It takes in information into a computer, then improves its performance as it goes. For example a search engine supplies you some results then you give it more information by clicking a result.
This is where a computer augments your intelligence, it takes something you have then improves it. An example he gives is where a deep-learning algorithm takes a picture you drew then you pass it to the computer and the output looks like a picture drawn by Van Gogh.
What it does is augment some ones creativity.
He gives other examples such a translation, it takes your words then feeds them to an algorithm then outputs the same words in another language making it seem like you have the ability to speak another language.
II, Intelligent Infrastructure
But I also think there’s something known as Intelligent infrastructure.
This is more where the world around us learns how to operate better. He describes it as things that take information from your phone say then orders a taxi and the taxi just takes you over to your destination. It’s where everything in the world works together to make things more smooth.
Conclusion
Basically what his views get at is that we’re far from AI, we’re currently working on the Machine Learning aspects to tackle the II and IA problems which can then hopefully be used as inputs to AI. I highly recommend giving the video a watch it’s quite a realistic view about where the actually field is. He gives all these problems like the fact that we can’t even label objects correctly yet from images.
Here’s the video if you’re interested: