The AI Can Now Make New Foods After Years Of Training

If all goes well, we could expect that the artificial intelligence will gradually take on a more major role in the development of our daily foods. For the moment, there are some big companies on the work with all the experts like head chefs, flavour scientists, and food experts to make a single AI that is capable of making food faster and better than human.

We not quite have the smartest algorithms yet we are getting there for sure. Like the AI lab of  IBM with their AI algorithm that is also cookbook-writing and Jeopardy-winning. This AI name is Watson and it has been collaborating with food developers from McCormick & Company for four years already. The first food products of this collaboration will be on shelves at the end of 2019, if all goes well, they expect to see all developer of McCormick to collaborate with the AI by the end of 2021.

According to Hamed Faridi, McCormick's science office it would take about 15 - 20 years to train a good product developer. So instead of that their vision is to develop the AI and the organization as good as the best developer in the world.

They hope for an end result of AI-made flavours, mixtures, and food based on knowledge of years of researching and learning on flavour profiles. Hamed said they are collecting a hundred thousands of formulas from the developers.

Follow the story of Hamed, he explained to us that as of currently to develop a new food product it will take them weeks or even months.

To develop a new product for McCormick, there are many specific detailed tasks that need the participation of a hundred developers. Sometime they would need psychologists, chemists, chefs and many experts in food all around the world just to develop a single product. They would take weeks to modify the old formula, add in and remove the ingredient, test ideas …

The AI of IBM would do the same tasks too but it includes deeper and broader knowledge about flavour and food than a single expert. They expect their system to design a product that is ready for the shelf in a fraction of the time it used to cost.

Normally a formula would need up to 150 iterations before hit the shelf, but with the new AI, they want to do it in just 50 tries with a normal developer. They are confident that they can reduce 70 per cent of the workload of the developer.

With the AI system of IBM, there is much more potential than just speed up the process. The system can also make formulas and flavours that human might not pay attention to.

It is natural that human developer has biases which are not bad in fact. The human developer will have favourite ingredients like a type of spices that will appear a lot in their formulas. On the other hand, the AI is just a massive neural network, loaded and trained with years of data. This system will be capable of considering all the alternatives much wider than just a single expert's wheelhouse.

The contextual data of the market research will be also taken into account. The AI's network of neural will process collected data on preferences like moods, location, and culture when designing a formula. With the contextual data, the AI made formula will be unique and surely will be valued by consumers.

The latter half of 2019 will be the time AI made foods hit the shelf. Some examples such as Tuscan chicken on seasoning mixes, sausage with New Orleans style, and much more to come. Moreover, according to Hamed and Goodwin, they have the plan to push the development of the AI system much further.

According to Goodwin:

The expectation of Hamed is that the collaboration of human-machine will change completely the development of foods. His vision is that of a system that human acts as the centre of the AI system rather than just all automated.

In addition, Goodwin visuals a future where AI system provides knowledge and humans focus on the creative part. For example, the AI could take care of the technical part like the combination or form of the ingredients while the expert could deal with the theme and the overall flavour.

The system is just like an apprentice, they do and they learn. It will gradually improve over time, of course not only the AI system of McCormick, but the soon to enter competitors as well. Hamed thinks that this will be just like electric cars.