SAIL: Two paths; AI as addon. AI as transformative
Welcome to Sensemaking, AI, and Learning (SAIL) a (twice?) weekly look at what's happening with AI and how that might impact learning and education.
The last few weeks have been heavy travel (Dallas, Phoenix, Seattle, Portland, San Diego). During that time, I've had the opportunity to engage with senior leaders in universities around their vision for AI. We're still early in the game, but there are two camps emerging. One views AI as another type of technology that will be adopted into existing university activities (such as advising, pathway guidance, guiding student success). In this approach, AI is primarily an agent for tweaking parts of the existing education system.
The second view is more radical and held by a smaller group of university leaders. In this view, AI is a catalyst (on top of many existing change pressures that the university already faces) that will result in a dramatically different future for higher education. Courses? They'll change. What we teach? That will change. Transcript? That will change. Imagine a future where profiles/learning graphs capture what a student knows and how she has come to know it, computed curriculum that responds in real time to advances in student's knowledge (linked to what we would today call a degree), and labor market skills evaluated and rapidly incorporated into academic programs. This triad (profile, computed curriculum, labor market relevance) is hard to enact in today's rigid university structure. A few innovative universities will begin experimenting here and will do so outside of the traditional system, likely with private partners. Those universities will be big winners in the emerging academic landscape. The benefit of first mover advantage, life time lock-in through profiles, and the costs of developing an AI-First University will reduce the number of universities. And it will accelerate the nascent globalization of higher education.
AI and Learning:
We're hosting two webinars in the next few weeks:
Building AI Applications Based on Learning Research. Hear how Khan Academy developed their AI Chatbot and important lessons universities can take from this experience. Register here.
ePortfolios and self-regulated learning: Promises, problems, and possibilities. Register here.
There are many options for improving research, especially around literature reviews. There is room for improvement, however. (while this is in Nature, not sure what's going on there. This reads more like a blog post)
Speaking of Nature blog posts, here's another advocating for the use of open source generative AI models.
An education/AI conversation with Bill Gates at the ASU-GSV Summit this week. AI will be as good a tutor as a human.
AI Technology:
A more visual way of interacting with LLMs. Looks interesting, but it's just for demo purposes.
HealthGPT: Connecting personal health data with ChatGPT. The rather obvious application would be a learningGPT system that would allow students to have discussions with all of their previous academic work. The concept is interesting. We have a large amount of data (health, travel, email, work outputs, text messages) that can be better accessed and engaged with through the use of a chat agent.
OpenAssistant. Everyone gets a conversational agent (see point above about a chatbot interface to our own information)
Microsoft is developing its own AI chips (or so goes the rumor). Given how much of the current deep learning work relies on GPUs, I imagine all big tech companies have been working on their own.
The future is visual and multimodal. Here is a simple concept Minigpt-4 (see video of the concept in action). Basically, chat with images.
Text-to-video is going to be fascinating. Nvidia shares their research. Scroll halfway down to see some examples.
Data, computation, and algorithms are the three things most responsible for our current AI cycle. Data, however, is about to get more expensive for model-builders. For human interaction, Meta likely has the best global data sets (FB, Insta, WhatsApp). Twitter and Reddit are also valuable. "Community platforms that fuel LLMs absolutely should be compensated for their contributions so that companies like us can reinvest back into our communities to continue to make them thrive"
Autogpt and babygpt are absolutely fascinating: “Autonomously achieve whatever goal you set”. Here's an excellent intro. Here are a few examples
The inside story of chatGPT's astonishing potential (a clearly humble title). But it is a good overview by an OpenAI co-founder. These agents are a new user interface to information and to getting things done. A good description of the role humans play in fine tuning models is presented. First 15 min is a presentation, then a discussion (starting with why OpenAI is outperforming the much larger Google AI talent).
Google is merging Google Brain and Deepmind: Announcing Google Deepmind "we live in a time in which AI research and technology is advancing exponentially. In the coming years, AI - and ultimately AGI - has the potential to drive one of the greatest social, economic and scientific transformations in history."