Artificial Intelligence, where will it take us?

March 28

This week I will focus on Artificial Intelligence. It is a growing field and there are many new developments even though it is early days.

Anything is Possible

Artificial Intelligence is only a series of software programs. How then, do we expect software to learn? There are multiple emerging tools. One popular tool is a Generative Adversarial Network or GAN. Introduced in 2014 by Ian Goodfellow, GANs are a further development of the work first proposed by Jurgen Schmidhuber in 1990.

Legend has it that Goodfellow and a few mates were at the pub one night in 2014 and were arguing about the potential for a how a machine could learn from itself. Goodfellow proposed the system that became known as a GAN. He bet his friends that he could make it work, went home and started coding and by dawn, he had won his bet. GANs were born.

Put simply a GAN is two competing programs. One is called a Generator and the other is called a Discriminator. The idea is that the Generator creates something, e.g. a drawing of a cat, and shows the Discriminator. The Discriminator has been taught what a drawing of a cat looks like and it judges the drawing from the Generator based upon its’ prior learning. Based upon the feedback given the Generator then tries again. the Discriminator gives feedback and the loop continues until the Discriminator approves the drawing from the Generator, as a drawing of a cat. Computers can work quickly and don’t need to stop for breaks. Learning happens very quickly.

So what use is this? Have a look at the following faces. They are not real, they have all been generated by a GAN. They have been created by adding face A to face B.

How would this be useful? Advertisers that don’t want to pay royalties or fees to actors for ads or movies that want to reduce the cost of extras might use them. Uses are just starting to emerge. Also scams based upon fake celebrity photos are proliferating. Don’t believe everything that you see.

To date, GANs have mostly been used in image generation or development however new applications are being developed in writing and scientific research. Designers are also starting to use them in design development. The potential is limitless.

From Success to Significance

One huge area of development for Artificial Intelligence is Health. There are many systems being developed however there is a significant stumbling block.

AI needs huge amounts of data to train systems. It can be difficult to obtain health data ethically (yes, that hasn’t stopped most tech companies but that will change). One way to solve this issue is called Federated Learning. It was developed by Google to read Android phone users text messages in order to improve predictive text. Yes, Google can and does read your text messages and email, go back and read the terms and conditions that you clicked “agree” to without reading. The system protects privacy by only accessing the message and not any other data about the user.

This approach is starting to prove useful in accessing training data for AI and health systems. The AI has no use for your name or address. The technique could be used to train AI on wider samples of data to help reduce and remove unconscious biases that emerge with smaller training sets. Large training sets allow AI to learn how to read Radiography Images and scans and to identify abnormalities for specialists to review. This article goes deeper into this subject for those that are interested.

Another interesting area of research is in developing diets specifically for individuals. This New York Times story follows a reporter who participated in the development of a personalized diet based upon the reporters genetics and reaction to individual foods over a two-week period. Blood glucose was continually monitored and stool samples were tested along with a meticulous tracking of everything eaten or drunk during the test period. AI was then used to combine the results from more than 1,000 other people and develop a recommended diet to help the reporter live a longer and healthier life.

There are a few online companies that claim to be able to develop a diet for you based upon a DNA sample but those claims are dubious. Deeper analysis is required. In the future it is possible that we will all have diets that are individually tailored to our needs. All we need now is for someone to invent a way for humans to stick to a diet.

From Intention to Innovation

One of the big problems that machines have is that they can not understand meaning in the same way as humans can detect meaning from the many nuances in our speech. Pat.ai is an Australian company that is trying to change that.

From their website “Pat Inc. is teaching language to machines. We deliver meaning-as-a-service: connecting natural language with structured information that creates meaning and understanding. Pat Inc. is building the world’s most powerful natural language service for developers to build intelligent agents and applications that you can talk to or text.”

The key is to base the language that the machine uses on linguistics. Pat does not use training data (as most other systems do) rather it allows the machine to learn by progressively understanding the way words are combined regarding real objects, people, processes and events in context. It does not try to learn patterns as most other systems do. Most systems struggle to match the language capability of a 3 year old.

John Ball the founder of Pat explains his approach in this video.

Pat is very much still in development and time will tell if they are able to achieve a level of understanding that other machine learning systems have not been able to accomplish.

image from Pat.ai

The Entrepreneurial Journey

An Australian and US start-up that is working in the area of AI is A-kin. Founded by Liesl Yearsly and Travis Giggy, A-kin is developing AI that will autonomously solve complex problems and form deep and trusted relationships with humans. A personal AI for each of us to help us as we need assistance. They have a deep focus on the development of ethical AI.

A-kin has 5 main research areas:

• Autonomous Problem Solving and Decisions

• Human AI relationships

• Ethical AI

• Optimization of AI

• Epigenesis Learning Feedback (multidisciplinary approach to learning merging human learning theories and Machine Learning techniques)

Liesl also founded Cognea AI that was acquired by IBM Watson and it is basically the front end of Watson today. Watson is IBM’s major AI intiative. Additionally she was CEO of Mooter Search (Applied AI) and founded TopTots Cognitive.

Recently Liesl gave a talk on the history, development and future of AI at a Firemark Ventures breakfast meetup (Firemark Ventures is the venture arm of IAG Insurance). The talk can be viewed on Youtube here. It is a very easily followed presentation and great as an introduction to AI and its uses in the marketplace. I would recommend it to anyone wanting an introduction to this complex and growing field.

Paying it Forward

If you have a start-up or know of a start-up that has a product ready for market please let me know. I would be happy to have a look and give the start-up a shout out.

If you have any questions or comments please email me via my website craigcarlyon.com I would love to talk to you about your startup and see if there is any way that I can help.

Till next week.

The Logo on the header is “Thinking by Gregor Cresnar from the Noun Project”