Has there been increased interest in artificial intelligence (AI) and machine learning from the mining industry in recent years?
I would say yes, in the last two years especially. At the last Prospectors & Developers Association of Canada (PDAC) conference [in Toronto, Canada in March] there was even a section dedicated to machine learning and artificial intelligence, which was the first time that happened. This is really something very recent for the mining industry, but it is catching up.
Why do you think it is it important for mining companies to get on board with these technologies?
It would be the same as if we had told mining companies 20 years ago, “Don’t use email, it will never be as good as writing a letter”. Can you imagine a company now, these days, that is not using email? If you don’t climb on this train, you’re going to be obsolete.
I have read a quote saying three out of every five executives think that if they do not apply artificial intelligence or machine learning techniques, they will become obsolete in the next two to three years. I really believe that if the mining industry doesn’t apply these techniques, they will be… Not obsolete, but they would be not as competitive as anybody else that is using these techniques.
Is implementing AI and machine learning less complicated than many people think it is?
Definitely, yes. Although I believe that the machine learning industry says: “This is so complicated, you will not understand this thing, just use it” – but that’s not true. Machine learning was created by humans and can be used by humans. They also say: “You will need to have a PhD to use it”, but I don’t have a PhD and I use machine learning all the time. So it is accessible to everybody.
There are big companies like IBM with products like Watson. Watson can play chess, can help doctors in diagnosis, etc. There is one company that I know of that is using Watson in the geological study of its data.
But the thing is that you really don’t need IBM Watson for machine learning – there are plenty of machine learning programs that are available for free on the internet, and you can start with those. Machine learning is very accessible.
On one side you have Watson and IBM, which is more expensive and a little bit out of the reach of a normal junior company. On the other side, you have open-source programs; for example, Weka, a program created in New Zealand, which is completely free.
The problem with that one is because it open source, you really need to do everything yourself. You need to know how to process the data, how to teach the machine to learn, so it’s a little bit more complicated.
Then in the middle you have a series of programs like RapidMiner, Talend, or BayesiaLab. They have created an interface that is very simple to use, but they are more expensive – you need to pay for that. So, you have options that ranges from free to thousand of dollars.
I recently wrote a book on machine learning where I explain how to work with Weka and other software. I have been working as a geologist and geochemist for 36 years, and geomathematical modelling is my speciality, but I am absolutely astonished by the results that you get with machine learning.
Do companies need Big Data to use machine learning?
No, and that’s another common misunderstanding. Because of the price of analysing Big Data, people were forced to invent machine learning. That doesn’t mean that you can’t use machine learning for normal-sized data.
For example, for exploration drilling, you might be drilling 10,000m. Let’s say that’s 10,000 data points, each of them with 36 elements, so that is 36,000 values. You can use machine learning to process that perfectly. Except, for example, there is the RapidMiner system, which requires a minimum of 100 samples; that’s the only program that I know of that requires a minimum number of samples.
But for others, like Talend or Weka, you can work with five data points if you want. The thing is that the more data you have, the better the machine learning process is, and the better and more unaffected are the results that you get from the processing of the data.
Machine learning is not statistics. It will use some statistical methods, but it has the ability to find relationships and links that are not obvious. Machine learning will help you process the data and track information that otherwise you would not get by normal statistical processing.
Also, machine learning techniques are not geologists, they just process the data. When you get the unexpected result from machine learning analysing the data, it is your job as a geologist to interpret it – there is always going to be the human factor.
What do you think is the future of AI, machine learning and data analytics in the mining industry?
I think the future is bright. But all the companies that were presenting or talking about artificial intelligence and machine learning this year at PDAC, they were all doing one step – they were teaching their systems to work.
That means that they were only feeding data to Watson, for example they sent it the results from all the exploration work done in Quebec. They’re just at that point in the process of feeding it data, and the priority was: “can you help me do a search within my data?”
For example, let’s say I would like to know the location of all the drill holes that got gold values of 3.3g/t and copper values of 10ppm. Of course, you wouldn’t search that by hand, as it would take you forever; even with normal statistics it would take a long time. However, machine learning will do that in seconds because that is the kind of algorithm it can construct. That is just the tip of the iceberg.
Because mining companies are using Watson, which is a huge machine learning system, they need to spend a lot of time teaching Watson how to do it. With the other systems I mentioned before, Weka or RapidMiner, you can start processing your data in an hour – it takes a little time to prepare the data and clean it, then you just put it in the machine, select an algorithm, and you start getting results.
I believe that nobody in the mining industry is using machine learning to the full potential… Unless they are not talking about it, they may be doing a lot of things but not talking about it publicly. I haven’t heard about anything besides the process of teaching the machine how to learn. Next year at PDAC [in March 2020], I am assuming that there will be a lot of companies publishing results.
I have been doing machine learning analysis for several companies for the last year and a half, and the results are just beyond this world. That’s why I want to go to the Future of Mining Americas conference and present these results and these capabilities; my main objective is for people to understand that machine learning is not only for the big companies. Any junior, even a consultant, can have access to this incredible tool. And secondly, that you really don’t need huge data sets, that you can use machine learning for 100 samples or 200 samples and you will get incredibly good results.
Join us at Future of Mining Americas on October 21-22, 2019, in Denver, Colorado, US, where Ricardo Valls will speak about ‘AI, ML and Modern Mineral Exploration: How to Start Today’, as well as sharing his expertise on the panel ‘Equality in Innovation’.
For more information on this year’s programme and speaker line-up please click here.
Credit: Google News