This Week's AI Developments that Impact Education
January 25, 2024
Photo of the new AI Industrial Education Model
TLDR
*AI in your browser
*Advances in text-to-video
* Significantly more compute is being poured into AI development
*Deep fakes grow
*Another frontier model that can reason
*AIs that can reason + AI agents = agents that can reason and will transform education
*Phones that can translate in real-time
As we say in debate, “A brief overview at the top.”
#1. As leading AI scientist and innovator Andrew Ng said this week, it’s hard for him to answer the question, “How will AI change things?” because AI changes everything. It’s at least as big as electricity, so imagine how electricity impacts things at your educational institution. AI will affect at least as many things. One difference is that running electric lines took decades, so we had decades to move from the one-room schoolhouse to “mass education.” Everyone now has AI instantly on their devices, so education will also change instantly.
And if AI doesn’t change things within education, it will just change things without education. Powerful teaching and tutoring bots are emerging. Will education try to compete with them or integrate them? I sense that educational institutions that try to compete with them will lose, and those that try to integrate with them will win.
#2. AI is already increasing inequality. Nothing makes that clearer than this LinkedIn post.
The AI digital divide is already here. There are students in all levels of K-16+ education who are learning from teachers and professors who are integrating AI to the best of their ability, and then there are those who can’t even access it at school or on a school computer. The difference is real, and I can see it. I can only imagine what it will be like in another year, as some continue to learn with pencils and others learn with computers powered by electricity.
Now for the developments.
Practical Google-powered AI in your browser. Google has introduced generative AI and ML capabilities into its Chrome browser, enabling users to select AI-created themes, manage multiple tabs efficiently, and receive AI-assisted writing support. These features are available in an experimental release starting today.
Implications: It’s getting harder and harder for schools to control AI, and it increases productivity access for those who take advantage of it.
New text-to-video. Google released (for demonstration) a new text-to-video tool.
PlayPlay is also coming for this market.
Implications: If your students learn video design, what tools are they using?
Voice and video avatars. Eleven Labs, which allows avatars to be generated from photos in minutes and can duplicate a person’s voice in seconds, raised $80 million. It can also capture a person’s facial expressions and tone when translating into more than 40 languages. It is being used by 42% of the Fortune 500.
Argentina’s President got a lot of press when he used this to translate his speech at Davos (click on the link to play).
While this shocked a lot of people, this is something I did four months ago with the same HeyGen tool.
Implications: Expect more deep-faked phone calls to your educational institutions. There are even reports of students using the technology to have the voices of their parents call and initiate conversations to call their child off sick. This is small potatoes in the grand scheme of things, but given the rapid duplication of voices with only seconds of voice detection, administrators should develop code words that only they know to communicate with each other in case of emergencies. Sam Altman has recommended this to families to prevent scams (AIs calling as people’s children saying they need money). On the positive side, schools can use this to instantly translate messages they want to get out to students and families in their communities who speak different languages.
Deep fakes. A digitally manipulated robocall, falsely portraying President Biden as the caller and discouraging voting in New Hampshire's primary, highlights the growing challenge of AI-generated content in elections, raising concerns about voter manipulation and misinformation. The incident is seen as a precursor to more sophisticated voter suppression and election worker attacks through AI technology. They may never catch the culprit, and even if they do, the damage is done and could be more significant in the future.
Graphic AI images of Taylor Swift and Travis Kelcie have now gone viral, creating a similar situation to what happened with high-school-aged girls in New Jersey and Spain this year.
AI literacy anyone?
Facebook emphasizes its commitment to artificial general intelligence (AGI). Last week, Facebook made a public announcement that they are trying to obtain AGI and they are working on the development of Llama-3. The media took this as something new, but Facebook has been working on trying to achieve AGI for quite some time; now they are being public about it. What is significant to me is not that another company is working on AGI, but that Facebook is purchasing another 300,000 GPUs for the effort, which puts them in the ballpark of OpenAI’s GPU usage. This means there will be even more significant advances in AI along the way to AGI. As Mark Zuckerberg notes in this article, AGI won’t be a magical moment but will be achieved through the development of more and more AI capabilities, so what this means is that more incredible AI capabilities will come even faster.
On a related note, Adept’s Fuyu-Heavy is now competitive with ChatGPT4 and Gemini Pro, even exceeding it at multimodal reasoning (Yes, these models are no longer simply regurgitating predicted text).
As I’ve said many times, the fact that models can do more and more math means their reasoning capabilities are improving.
Implications: The enhanced reasoning abilities in AI are significant because it means that AI can offer more personalized and effective learning experiences. AI that can reason well adapts teaching content to the unique styles and needs of each student, identifying their strengths and weaknesses for targeted learning. This leads to an educational environment where students receive the exact kind of support they need. Furthermore, AI's reasoning capabilities enable the development of simulations, providing students with practical, hands-on experiences that deepen their understanding of complex subjects.
Alongside this, autonomous agents are becoming an integral part of the educational framework. These AI-powered assistants can independently interact with students, offering real-time guidance, answering queries, and providing interactive learning activities. They act like virtual tutors, enriching the learning process through personalized engagement and allowing teachers to focus on more nuanced aspects of education. Together, AI's reasoning capabilities and autonomous agents will transform education into a more adaptive, interactive, and future-ready domain.
AI comes to phones. AI is running locally on the new Samsung phone, enabling live, real-time language translation and new levels of instant image searching and referencing
Apple is currently working on a phone with similar capabilities, including making it easier to run AI models locally on the phone.
Implications: Student readiness for just-in-time learning will expand rapidly, and having them seriously engage in learning a language will present a greater challenge.