AI is Crucial in the Technological Race Between East and West

In the past, AI has displayed that it can help in research. An example of this is D3M, which featured on AIDaily last week. However, as technologies improved, so has the potential uses of AI. Recently, machine learning techniques for handling words, also known as "natural language processing" have improved vastly. Large corporations have recently taken advantage of this and improved their assistants.

However, this improvement isn't restricted to large corporations alone. A group of scientists at Berkley Laboratory decided to run an artificial intelligence model, called Word2Vec, on a dataset of research papers, training the model to detect potentially thermoelectric materials. The group went onto publish a study on July 3rd to the Nature publication.

Thermoelectric materials are materials that convert electrical energy to thermal energy and vice versa. An example of a thermoelectric material is Bismuth Telluride (don't worry if you haven't heard of it - it's not a household item). It is commonly used for converting heat into electricity for refrigeration or for creating a portable power generator.

Passing enough heat into a thermoelectric material will cause a potential difference and current to be generated across the material, making it essentially a battery that runs on heat. No world changing candidate has been found so far, but maybe that can change if AI is used more and more.

Passing enough heat into a thermoelectric material will cause a potential difference and current to be generated across the material, making it essentially a battery that runs on heat. No world changing candidate has been found so far, but maybe that can change if AI is used more and more.

It may seem complex to someone who has never studied material science, However, do rest assured, it is fairly complex and the AI was still able to make discoveries. This is due to AI recently being trained not to "know" stuff as may be expected from artificial intelligence, but to notice differences and classify things that are different separately in memory and add to the list as soon as it detects something in those groups. An example of this would be showing AI models pictures of cats and dogs. The model would output 2 groups that it has recognised differently, but it wouldn't have the faintest idea of what a cat or a dog is.

In a similar fashion, the model read a whopping 3.3 million studies relating to thermoelectric materials and learnt how thermoelectric materials are described in papers. After that, it was fed other research papers investigating different materials. The model, having been fed descriptions of many materials, then attempted to find the ones that matched the descriptions of thermoelectric materials; and it did that successfully.

Jain, a researcher on the team, claimed it could make connections in research papers that "no scientist could". With many "cross-discipline associations". It was so effective, it managed to sensibly link words that were never even mentioned in the same paper but simply because the machine was able to understand that they described similar things.

To truly check if their model words, they dialled back time all the way back to 2009 and put research papers up to and including 2009 into the model to try and discover materials that were discovered much later, and the model succeeded again. It managed to detect one of the most effective thermoelectric materials 4 years before scientists discovered the material was an effective thermoelectric material.

The main point to take home is that AI did not discover anything, the model doesn't control a lab (though they may in the future), but instead, it managed to piece together a vast network of information in a way no man can do. There are simply too many research papers being published for someone to read and some materials fly under the radar. While this is hard for humans to do, this is a perfect job for an algorithm. Once more, machine learning algorithms save the day in research.

In light of recent events such as Russia and China heavily investing in both AI and research, it has become more and more apparent the western world must also invest in AI. However, this investment shouldn't just be in corporations using AI, but artificial intelligence to advance research too.

China's main technological sector, also known as the silicon valley of the east, has advanced the nation's scientific prowess by a sizeable amount. Ranging from hiding cameras behind LED screens so you can finally get rid of the wretched notch to headsets that read brainwaves to guess your emotions. It is essential the west doesn't fall behind and investing in AI in research may just help the west get a leg up in the race for increasingly advanced technology.

Parth Mahendra