Today I’d like changes things up a bit and talk about the future. Specifically, I’d like to talk about artificial intelligence (AI). Like electricity, computers and the internet, AI is technology that is very disruptive. Disruptive technology is something so different that it can change the way a business operates and in the extreme can transform industries. For this article, I will draw on a lecture presented by Andrew Ng, Baidu chief scientist and adjunct professor at Stanford University. The lecture was made to business students at Stanford early this year and can be found on YouTube here. We give full credit to Andrew Ng for many of the thoughts in this article (including the title of the post). On a personal note, I have being researching machine learning applications for financial market timing. I was in high tech for 25+ years and the rate of progress I am seeing in machine learning algorithms seems to be as fast as any technology I have seen (computer adoption, networking, wireless networking, data communication & storage, and graphics processing come to mind as other very notable shifts).
What is Artificial Intelligence?
Artificial intelligence is a term that conjures up notions of computers that have the ability perceive, learn and reason on par or superior to human brains. There are many AI techniques and one thing that helps us to think about AI is to focus on the current state. Where is AI being used? What problems can it solve? In his lecture, Mr. Ng focused his discussion of AI on machine learning algorithms. This is a subset of AI and an area that has seen explosive growth in applications in the last 5-10 years. For this discussion, let’s focus on what machine learning and it’s current and near future outlook.
What is Machine Learning?
Mr. Ng describes supervised machine learning as a process where we take some input and generate an output. For example, input could be email and the output is a decision as to whether the email is ‘spam’. Computer scientist might further break down AI into two distinct categories: classification and prediction. In classification, we attempt to take some input and determine what class the item belongs to. ‘Is this image an image of a fish’ is a classification question. As an example of prediction, we might ask how many board feet of timber will a forest produce in ten years given certain conditions. One key element of machine learning is the idea that the algorithm takes in data and is able to ‘learn’ from the information. To understand learning, consider the example of spam filtering. A machine learning algorithm may be fed millions of emails including whether or not those emails are ‘spam’. The algorithm uses this data and builds a model for ‘spam’. Now, when I feed it new emails the model, having ‘learned’ about other emails that are spam, will classify the new email as ‘spam’ or ‘not spam’.
Just Exactly How Much Intelligence Do Machines Have Today?
Because the concept of AI is so broad in scope, it is helpful to have a rule of thumb to understand just how much intelligence machines have today. According to Mr. Ng, he suggests that today’s machine learning is capable of about as much ‘thought’ as a human brain can do in 1 second. Wow! This helps us put AI in context. From the state of AI today to a fully sentient machine (think Hal from Space Odyssey or The Matrix).
Where is AI / Machine Learning Being Used Today?
There are many examples of where AI is in use today. It is already transforming industries. Here are a few examples. Image recognition is a major area of application. For example, the latest iPhone introduced by Apple includes the ability to determine whether the image that it sees from the camera is your face. It can use this information as a security mechanism to unlock the phone. The feature requires both software and a ‘neural engine’ hardware component built into the A11 processor on the phone. The neural engine is capable of running machine learning algorithms faster than general purpose software.
Advertising has already been transformed by AI. Mr. Ng gave the example of web based advertising. Given past data from users and their profiles, Baidu can predict whether a user is likely to click on a particular ad. Advertisers pay based on whether a user ‘clicks through’ and if Baidu and other advertising based companies such as Google can improve the clicks their revenues go up dramatically.
Speech recognition is another area where machine learning is in use. Baidu has thousands of hours of audio in many languages. They can use the data to ‘learn’ how to convert the audio to text. For example, in e-commerce you could use this information to allow online shopping using voice commands.
Recently my son and I traveled to Stanford from San Jose via Uber. We learned that we were paying less for the ride then the driver collected. What explains this? Uber is using AI to determine rates based on what consumers will be likely to pay given time of day and route. Using the technology it could be that Uber is focused on maximizing utilization to gain scale. Speaking of driving, self-driving cars are a hot new area of development. Tesla is working on electric self-driving semi-trucks. We are a few years off from seeing wide use of self-driving vehicles, but I wouldn’t bet against them being here in 10 years.
Google is already using AI in Google Maps. For example, AI is used to censor out license plate numbers and faces. It can also detect information from street signs. AI is used for routing drivers in the most efficient manner. There are many more examples of where AI is being used. In fact, it’s hard to think of an industry that isn’t using AI today.
Why is AI Growing so Rapidly?
There are several reasons for this. Mr. Ng points out two major factors: the amount of data we have and the amount of machine learning performance we have. In the first case, it is easy to recognize that we are awash in data. This has been enabled by the internet, computing and massive amounts of storage. We went from written mail to email. We have gone from store purchasing to online purchasing. We went from paper maps to online maps. Newspaper, what’s that? The print advertising industry has been in a multi-decade decline as advertising dollars have shifted massively to online forms. All said there are very few places where we don’t have online data.
As for processing power we have a few factors. First, we have a huge amount of raw processing power on computers. I have a 4-core Intel I7 processor that runs machine learning routines on a gigabyte of data in minutes. Aside from the most complex machine learning problems, we have enough computing power in our phones and PCs to do a heck of a lot of work. Now, on top of this there have been rapid advancements in the state of software. I have open source software on my home machine that is capable of deep neural network generation. All open source. Machine learning software power coupled with high computing power is enabling the wave.
What are the Social and Ethical Issues Involved?
First, at the core, we are automating tasks humans used to do with AI. As a result, jobs will be lost. This is not unique to AI. Technology in general boosts productivity (more output per human), but comes at the cost of losing jobs. Certain industries will be impacted more than other industries. One example Mr. Ng gave was Medical Imaging Technicians. These people review diagnostic images and make determinations such as ‘does this person have a tumor’. These types of determinations can be made by machine learning techniques.
The Ethics of AI have also come under scrutiny. Just this morning, I watched an interview with an ex-Google marketing manager where he discussed the topic on CNBC. He acknowledged that using AI, companies have the power to ‘shape’ and ‘focus’ where people spend their time and money. Companies building the technology say that people benefit, but there is an inherent conflict of interest based on the profits they are driving.
Finally, there are concerns about the long term risk of AI to the human race. Sounds like a science fiction movie, but the idea is being openly discussed. Recently, Elon Musk, Tesla CEO, commented that we need to be aware of potential dangers of AI. Mark Zuckerberg, Facebook CEO, called his comments ‘irresponsible’. Mr. Ng’s point of view discussed in the Stanford talk is that sentient AI is a long way down the road. I don’t think that AI danger should be in the top 20 things that keep us up at night, but being aware of where AI is being used (for or against us) seems prudent.
Now that we have a good grounding for what AI is, I hope to do a follow-up article that goes into a bit more discussion on how it may impact investing going forward.