This post was originally published by at Medium [AI]
Machine learning is an excellent assistant for inventing new products
How do you invent a new perfume? A new dish? A new flavor of whiskey? While these products may seem wildly different, the process for creating them can be described vaguely as learning what parameters go well together and then combining them. For example, a perfumer may have about 1,300 substances available to them when crafting new fragrances. By combining some amount of these substances in new ways, one can discover a new scent.
Incidentally, this is a perfect job for a machine learning algorithm. Feed an algorithm with information about these substances, along with formulas of past perfumes, and it too can begin to predict new scents. Human experts can then examine and refine these AI-predicted recipes.
Allow me to introduce you to three companies and one duo of academic researchers who, in this fashion, created new products using machine learning algorithms. Mackmyra crafted a single bottle of whiskey together with AI, while both McCormick & Company and Symrise continuously create new products with AI. The former: food products. The latter: perfumes. Finally, our duo of researchers used machine learning to create a flu vaccine that outperformed existing vaccines.
Mackmyra used AI-human-collaboration to create a whiskey
In 2019, Mackmyra partnered up with Microsoft and became the first company in the world to create whiskey using artificial intelligence. By combining a substantial number of data points, the distillery’s machine learning AI was able to generate more than 70 million recipes, naturally sorted by likelihood to be of high quality. The AI could rapidly calculate which flavors went well together, and was also able to suggest innovative combinations that otherwise would have never been considered.
The recipes generated by the AI were analyzed by human experts who put their own experience and input into the process. This Sweden-based company released its first AI-powered whiskey in the autumn of 2019. They called it “Intelligens,” which, in case you needed a translation, is Swedish for intelligence. A fruity single malt whiskey, the company describes the whiskey’s flavor to contain notes of toffee, creamy vanilla, pear, apples, white pepper, and a light tone of toasted oak casks. Well, it must be a decent combination of flavors, seeing how this recipe beat 70 million others, no?
McCormick & Company creates new food flavors with AI
McCormick & Company, one of the largest food corporations in the US, has partnered up with IBM to create new food flavors using AI. Rolling out operations in at least 20 labs in 14 countries; the project aims to deliver new flavors for hundreds of products. The project has already resulted in unique discoveries, such as new spicing blends.
Creating new flavors is incredibly challenging. Product developers must sort through thousands of available ingredients, determine which go well together, and in what ratios. Yet product development is a vital method of establishing a competitive advantage.
There is an overwhelming amount of data available, making machine learning an excellent choice for the task. McCormick aims not just to use the AI to create new variations of existing flavors, such as a unique variety of vanilla, but flavors that have downright never been experienced before.
Symrise uses AI to create new perfumes
Symrise, a major producer of fragrances and flavorings, has begun to produce fragrances using machine learning. They created an AI named Philyra, which is working as an assistant to the company’s professional perfumers. While the AI naturally cannot smell fragrances, it can certainly piece them together. See, crafting perfumes is quite similar to building dishes, which is perhaps why this company also partnered up with IBM.
A pleasant-smelling perfume is formed by combining a set of substances. A perfumer has about 1,300 substances available to them. By giving Philyra access to these substances, along with a database of nearly 1.7 million perfumes, it too has learned which elements complement each other. The formulas it creates are even made a specific target audiences in mind. After Philyra has crafted a recipe, a human expert refines it.
However, in a test performed for a jury, a fragrance that was 100% AI-generated was overwhelmingly chosen as the jury’s favorite, rather than perfumes where humans had been involved in the process.
Researchers at Flinders University created a flu vaccine using AI
Developed by Australian researchers at Flinders University in 2019, an AI named SAM (Search Algorithms for Ligands) was constructed to create potent vaccines. The researchers first taught the AI to understand compounds. They allowed SAM to learn which compounds are able to activate the human immune system, and which aren’t. A separate system then generated trillions of compounds that were fed into SAM, which was given the mission of determining which of these trillions of chemical compounds might be good human immune drugs. Out of the pool selected by SAM, researchers tested the top candidates. The researchers discovered that SAM had not only identified the best alternatives but had even come up with better drugs than what currently exists.
The result was a vaccine against the flu which outperformed all other existing vaccines. This was not only the first vaccine developed by an AI but was also superior to any vaccine made by a human. The researchers argue that SAM can save millions of dollars in research and can cut development processes by decades in the development of future vaccines. Speaking of future vaccines, some organizations are indeed using machine learning to help develop a vaccine for Covid-19, such as Baidu and MIT.
There’s more. IntelligentX uses AI to make beer and NotCo uses AI to create plant-based alternatives to animal-based products. Human-AI-collaboration is, I think, an exciting next step in product development. I hope that using machine learning to empower human experts in creative roles will become the norm in the future. AI can help speed up the development of innovations, and it could even be used to help customize products tailored to specific groups of users.
There are many AI strategies that one can adopt to create value using machine learning. Finding the right strategy for the right business model is key.
Now, if you’ll excuse me, I’m going to go tinker on an algorithm that will come up with new cocktail recipes for me every weekend. Making the best of quarantine life, one cocktail at a time.
This post was originally published by at Medium [AI]