This special podcast is brought to you by Barclays Africa. Hod Lipson is with Singularity University. Heâs the expert on faculty, for AI and Robotics. How did you get involved with Singularity in the first place, Hod?
I am always interested in pushing the limits of our technology. AI, 3-D printing, and robotics. It doesnât take long before you stumble across Singularity.
How so? Where were you before?
I teach at Cornel University and Columbia University. Iâm involved in giving talks in many places on this topic.
Have you been to South Africa before?
I have been here only once, in the context of 3-D printing, actually â a couple of years ago.
You were billed as the 3-D printing expert. Just to unpack from a South African context (because itâs a very relevant issue here), we are going through an industrialisation program, which looks a little similar to what might have been thought about 20 years ago. 3-D printing could make that irrelevant. Things like putting up big factories and lots of unskilled labour.
3-D printing is not going replace mass production. The technology definitely changes disruptive to conventional manufacturing, but only certain types of manufacturing so I wouldnât say that in the short-term, itâs going to replace conventional manufacturing. It does open up new opportunities. The aerospace industry as well as the medical industry is adopting 3-D printing in big ways. If youâre making toothbrushes in factory for making commodity items at large numbers, then 3-D printing is not that disruptive so you have an entire range. It depends.
You guys at Singularity University talk about exponentiality. If 3-D printing continues developing the way that it has developed, surely, I might have one in my home one day and do my own toothbrush.
Thatâs right. If you take 3-D printing to an extreme, it is optimised and customised products. Instead of having a âone size fits allâ that isnât particularly good for anybody. Specifically you print parts on objects and theyâre optimised and customized for whomever needs it. Indeed, thereâs no reason why you wouldnât have a toothbrush thatâs perfect for you. I donât think youâd print it at home. I still think youâll print it somewhere else and have it shipped, using a drone or any of the other numerous technologies weâve been discussing. You donât necessarily need to have a printer at home but it could definitely unfold that way.
What can it do at this point? What can a 3-D printer actually produce?
Currently, 3-D printers can produce any shape and any part out of almost any material from plastics to metals, titanium, stainless steel, ceramics, and even biomaterials and food. Thereâs an entire range of materials. If youâre talking about whether you have a printer at home, my guess is that you will have a printer at home but it will be food printer and not a metal printer because food is the only thing you want to print and consume immediately whereas if you need a metal doorknob, you can wait until tomorrow.
That is mind-blowing to me that you can print your food at home. Where do the inputs come?
Just as with any printer, you have cartridges of material. These cartridges could be frozen, etcetera. You pop in maybe two dozen different materials â basic ingredients â and you download the recipe. The printer combines the basic ingredients together (like a print) and possibly cooks it in line with a laser or something else, and you have your food. Itâs not necessarily the way to print a steak and a salad, but any processed foods can be made that way.
How far away are we from seeing that in households?
Are you talking about food printing specifically? I would say the big question around food printing is more of the business model rather than technology. The technology is here. Weâve demonstrated it in the labs. A couple of people are working on it. The question is, âdo you charge for the recipe? Do you charge for the cartridges? Who makes the machines? Whatâs the business model around us?â Weâre in a bit of a âchicken and eggâ situation where we have machines, but no recipes. Itâs a bit like having iPods with no music. Itâs difficult to see how this thing kicks off. I think that in the next decade or so, youâd definitely think about food printers as yet another appliance in your kitchen.
Incredible. You spoke in some detail today about artificial intelligence and how itâs a combination of algorithms and data. Just unpack that.
Often, when we talk about AI, you think of a monolithic thing that makes decisions, but if you look a little bit inside, thereâs always an algorithm and thereâs always data. We have more and more data coming in from everywhere. From sensors, phones, biometrics, and satellites. We have better and faster algorithms, better-designed algorithms, and algorithms we never had before, and that combination of data and algorithms is really, what powers the AI revolution thatâs happening.
Just explain it, maybe in laymanâs terms: something where AI has been applied and where itâs used algorithms and data. Tell me a bit about Watson â IBMâs supercomputer.
Itâs difficult for me to describe in detail because I donât know exactly, how it works. Actually, thatâs a proprietary technology to IBM. You canât read a paper about how Watson works. However, if you look at most machine-learning techniques available â letâs say for predicting retail â you have a lot of data about how many items you sold the last quarter. You might have day-by-day information. You might have a lot of information about how many people were there, the weather, or what the stock market did that day. They have a lot of data about what is going on every day and you have your product sales. You can take all that data and you can have a computer algorithm look at that data and try to find a relationship between the number of items sold and all the rest of the data. If you find that relationship, automatically you can take it and you can make predictions. For example, how many items youâre going to sell next month, whatâs going to happen if itâs going to be sunny next week? Should you have more of these or less of these? You can start making much more educated decisions about the future of your factory or your store, better than what you would do if you were just doing it with gut feelings.
You donât have to be a PhD anymore, to do that. You were talking in your presentation about something, called Eureka.
Eureka is a spinout, out of our lab â one of these technologies that take a bunch of data and you can ask questions about it. You can say, âTell me the relationship between this item, which I care about, the number of items sold, the revenue, and all the other data items Iâve collectedâ. Eureka will sort through that and give you the answer, and itâs very easy to use. Itâs almost like a spreadsheet. If you have the data and you can put it in a spreadsheet format, you can go right ahead and use this kind of software to ask questions. You donât need to understand AI. You donât need to understand how it works.
How do you know what questions to ask, Hod?
Thatâs the real power. Where the human comes into all of this is collecting the data, figuring out what data is useful, and asking the right question. I keep emphasising that one thing. AI is now at the point where itâs an incredibly powerful tool for predicting the future, but you need to know what to ask. Itâs a little bit like an oracle. An oracle sits there and you have to ask the question. If you ask the right question, youâll get an answer thatâs useful and you can act on that result. If you ask the wrong kind of question, there might not be any coherent answer. Very often, the AI will say, âThereâs not enough data to answer your questionâ. If you ask, âWhatâs the winning lottery number tomorrow?â you havenât provided any data to answer that and the AI will tell you that.
Unless you give them the winning lottery tickets of the last 50 years.
Even then, if itâs a good lottery, there would be no correlation and the AI would work on this for a long time and say, âI cannot predictâ.
Project Cassandra.
Thatâs currently an ongoing project. Weâre trying to tackle the ultimate question, which is âCan we predict earthquakes?â Iâd say that earthquakes are the holy grail of data mining and prediction. Theyâre very difficult to predict. Theyâre very spare. Unlike the weather or the stock market where you collect a lot of data all the time, earthquakes are rare. Thereâs a long coessence with nothing happening, so thereâs a burst of activity. Taking all that data and trying to make a meaningful prediction on it, is very tricky. Weâve been working on this now for over a year, in trying to make headway in that space.
Just explain the kind of data that you put together. Does it go back hundreds of years and all over the planet?
Weâre using publicly available data. This is the U.S. Geological Survey data, so this basically, a recording of every earthquake (large or small), including ones we wouldnât even notice across the entire globe, since 1970. That data is public and weâre basically, using a very raw approach. Weâre saying, âForget about your physics modelling. Forget about prior knowledge about how the earth mantle works and plate tectonics. Enough talking about that. Weâre not even going to think about it. Weâre just going to take the raw data out of thisâ. Weâre going to put a big list of events into a big data-mining algorithm and see if it can find patterns. Itâs finding patterns and making predictions.
Thatâs scary. What do you do with it?
I think weâre at that point. Weâre not even sure how to publish this. On one hand, if we publish it and itâs right, weâve saved many lives. If we donât publish it (and it was right, we have blood on our hands. If we publish it and it doesnât happen, then we could cause a lot of economic hardship and unnecessary evacuation. Itâs a very tricky position to be in and I think this is partially why we called it Project Cassandra. Thereâs a little bit of a curse associated with knowing the future.
How confident are you that these earthquakes will happen?
This is why weâre not yet ready to actually release dates. We want to triple-check these things. Weâll never be 100 percent confident, but we want to be able to say, âThis is the confidence level. Here are the dates. Here are the locations and hereâs the confidence. You make your own decisionâ.
In other words, it could be a ten percent of 50 percent risk of it happening.
Itâs not a prophet who says, âThereâll be an earthquake on this date.â Itâs going to be around this week, within 100 miles of this location with a probability of 90 percent â something like that. At that point, Iâd say Governments and other experts need to weigh in and look at the data, look at the actual reasoning that the AI had, and say, âOkay, this makes senseâ or it doesnât. Unfortunately, sometimes itâs going to be very difficult for people to understand â even experts. In the end, itâs going to be difficult to make the call, whether you evacuate people or not. Iâm not sure how to deal with that.
Artificial intelligence and robotics. Hod Lipson is in South Africa with Singularity University and this special podcast was brought to you by Barclays Africa.