One of the fun volunteer jobs that I have is talking to American Red Cross volunteers about the wonderful work they do to help others in need. I recently wrote this post for the local chapter’s blog about the efforts of Dianne Tattitch (who works in IT for Mastercard) and what she did for those impacted by Hurricane Ian in Florida. Here she is helping with her guest’s laundry needs.

Once again I had an opportunity to judge several collegiate entrepreneurial efforts as part of the Enactus 2022 world cup competition. I was a virtual judge at last year’s competition and wrote about my experience then. This time around I was working with four teams, each of which had rather innovative ways to make and sell cattle feed. Now, such a mundane topic you wouldn’t think much of, and you would be wrong. I didn’t get to judge the ultimate winner, a team from Egypt. But I was impressed with the Tunisian team from the Higher Institute of Computer Science of El Manar. You can see their Enactus Report document here. What was impressive about the Tunisian students was how focused they were on solving several problems with their venture. First, they wanted to eliminate the use of imported corn and soy feedstocks that were very expensive for the ultimate feed product. They wanted to make use of by-products for human food production that could be used in animal feeds, and increase the nutritional value of the feed to provide better health and muscle production. Their project generated a net income of US$25,000 with 40 farmers using the feed that was a third cheaper than the existing commercial feed, produced entirely with Tunisian sources. They have plans to expand their project to neighboring countries next year.

I have been writing about authentication when it comes to the digital world for many years now. Last month I looked at authenticating sports memorabilia. Today’s story takes another look, this time about the ability to authenticate a painting by a world-class artist.

I got interested in this issue after reading a piece in the New Yorker about paintings by Lucien Freud (a relative of the doctor). The article mentioned a Swiss tech company called Art Recognition that uses machine learning and neural networks to authenticate art. I spoke to two of their 11 employees by phone to learn more about their technology and their customers, Ludovica Schaerf, an AI developer and data scientist and Romanas Einikis, their CTO and one of their founders.

Before the Swiss data scientists got involved, art experts required the actual artworks to be present in their own labs. This meant that the art had to be insured and shipped, typically great distances and at great cost. Once in their possession, the experts could keep the work for weeks or months as they examined it. “We don’t require the physical artwork in our presence,” said Einikis. “That saves on insurance and transportation costs, and also reduces the amount of time to obtain an analysis.” The scientists just require photographs of the work, and typically take a week to produce their analysis. A simple certification of authenticity costs less than $1,000.

The data team collects as many images of the artist’s work as they can obtain, typically from public domain sources or from museum and collectors’ websites. (This is legal under Swiss law, BTW.) These pictures — along with known fakes and similar work from other contemporary artists’ paintings  — are fed into more than 30 different data models that are run to produce a probability score. The models take advantage of cloud computing from AWS and Azure. “It doesn’t make sense to have on-premises machinery – it is a big headache and not worth it and the cloud is much more cost-effective,” he said. The models make use of NVIDIA GPUs and the CUDA tools for computer vision that were originally developed for video gaming.

To date, the company has found about half of the art works are fakes, which isn’t surprising given that the company gets called in when their provenance is questionable.

I asked a friend and former art gallery owner what she thought about this approach. She said that the art world is highly political and the traditional experts often have a vested interest in not being convinced by any computer program. She was concerned that many artists’ early works or unfinished works make this type of approach more difficult, but Einikis assured me that their models take this into account, along with incorporating information about which paintings were most likely created as collaborations among several assistants to the named artist. He mentioned that Reubens had periods of his painting career when his workers were helping paint the paintings. “We have to separate these different periods as part of our modeling process,” he said. So far, they are the sole provider of this type of service. It is an interesting intersection of art and science.