Béatrice Moissinac, PhD
Hello, world! 👋 Welcome to BeaBytes.
My goal is to help you understand AI and equip you with enough conceptual (but not technical) fluency to fight off the snake oil merchants.
I reserve the right to change my mind at any time.

contact -at- beabytes -dot- com

© 2013-2024 Béatrice Moissinac, all rights reserved.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of my employer or sponsors.
Can AI do Good?

Can AI do Good?

At its root, the power of Artificial Intelligence (AI) systems can benefit Humanity through its ability to automate decision-making in complex contexts, at scale. The quick rise of Large Language Models (LLMs) like ChatGPT has brought to the forefront questions about positive and negative externalities. Beyond the glamour and glitters of ChatGPT and beyond the hype and marketing ploy of “AI everything”, in this article, I want talk about the tangible positive Good that is produced by AI systems in the world today. 😀

AI can’t fold my laundry, but it can fold proteins

The application of AI in healthcare is not new. In Radiology and pathology, image recognition has been used to detect anomalies on mammograms and help support oncologists, and for example, help detect breast cancer earlier.

As technology advanced, so did many’s ambition. In 2012, IBM and Memorial Sloan Kettering (MSK) initiated a partnership to build a treatment recommendation system using MSK’s data. Ultimately, IBM Watson Health failed and was sold for parts, in part due to high maintenance cost.

One of the most exciting scientific advancement using algorithms is protein folding. Very simply put, protein folding is the process by which the different parts of the protein molecule will position themselves (fold, like an origami) to take on a structure that is biologically functional. From an algorithmic perspective, not all positions (folds) are pertinent, and we would have to “try them all” to figure out which one would work. Thus, protein folding is an expensive problem to solve. Using simulations, AI helps efficiently search and test the possibilities without having to make the molecule in real-life. In turns, this helps making faster and better new drug discoveries. This technology is vastly used with projects such as AlphaFold, EquiBind, or ATOM. You can even donate your idling computing devices to help open source projects!

I am usually very conservative in my assessment of LLM’s performance due to accuracy and ethical concerns, but I am genuinely excited to see what they can do with new molecule generation. If you want to learn way more about AI-assited drug discovery, there is this great talk from the Alan Turing Institute. 🤓 🧬

Prof. AI will be teaching you.

Throughout the years, algorithms and data have been used to construct Intelligent Tutoring Systems (ITS). ITS cannot (yet?) replace teachers, but they can provide support and assistance, by aiming to create an adaptive teaching/practicing experience. Mathematically, an AI tutor is trying to find your zone of proximal development, that is, to find an exercise, an explanation, etc, that is not too hard, and not too easy, to push your learning forward.

Some have been vastly successful, such as Duolingo and Khan Academy. They even publish their research openly. What does AI research look like at Duolingo? Take a look!

And some, like Knewton Alta (previously Knewton), have been riddled with the classic problem of over-promising and under-delivering. Moreover, a lot of applications of AI in Education are hidden within the service offering of large for-profit publishers, such as ETS AI Labs, Wiley, or Pearson.

But the advent and democratization of LLMs has push the capability of AI tutors even further. Start-ups like Merlyn Mind are developing a classroom AI assistant, which can reduce stress and save time. 🧑‍🏫 🧑‍🎓

AI against pollution & climate change

The negative environmental cost of LLMs is being documented with every new GPT model, and rightly criticized given the frivolity of the results.

Nevertheless, some class of algorithms can provide serious help to fight pollution. The Ocean Cleanup Project uses image recognition to detection plastic in the ocean. The Grey Parrot project also use image recognition to increase the amount of resources being recycled.

Most notably, the United Nation is actively sponsoring multiple research efforts to develop algorithms related to development goals. For example, weather prediction has been using numerical models that are very slow and expensive to run. AI algorithms have been shown to have the potential to reduce the computational time and resources needed to make weather forecasting.

They don’t always make the news, but risk management agencies and emergency services are already using AI research and algorithms to optimize resource management and improve detection, as extreme events are more and more likely to happen. For instance, wild fire detection in Washington state, as well as wild fire management. It has also been used to calculate whether Emergency response resources are properly allocated for optimal response time.

What’s next?

There are a lot of valid and legitimate concerns regarding the economic, societal, and environmental (to only cite a few) risks caused by the use or misuse of AI systems. Progress for better or for worse seem ineluctable. So every time I feel overwhelmed by the bombardment of bad news, I fight the media’s bias toward negativity by searching for “AI For Good” projects. I have pointed at some examples in this blog, but there are many more!

AI has terrific promises for the future and betterment of all of Humanity. It is our responsibility to make sure it delivers a cost-effective, fair, and equitable future for all.

NB: This was a feel-good post to balance next’s week post: A Quick Guide to the AIpocalypse. 😇