Stephen Downes

I only wrote a short contribution to this volume but I think the one-paragraph excerpt from it is a zinger: "Things will be smarter than we are: Instead of devising 'human-in-the-loop' policies to prevent AI from running amok, we will devise 'AI-in-the-loop' policies to help very fallible humans learn, think and create more effectively and more safely." I go on to discuss the 'second Copernican revolution' we are going through as humans contend with the realization that they are not at the centre of the thinking world; rocks can think. The full report is available from Elon University (286 page PDF).

As we all stare at the spectacle of the richest country in history complaining that 'everyone is taking advantage of them' it's worth pondering why the 'U.S.-made iPhone' is pure fantasy. "I am also baffled when I see Walmart selling a pair of jeans for less than $20. The only way this is possible is at a huge human cost." Not that anyone in the U.S. gov ernment is worried about that cost. "Workers' rights are not what U.S. tariffs are ultimately about. But the exploitation is ours. We, the richest countries in the world, go into developing nations and extract from their people and environment the products we want for the incredibly generous lifestyle we have." China is finally pushing back a bit against this exploitation, so Apple (and others) are just moving to India, and the beat goes on. 

"For the last few years," writes Alfie Kohn, "we have witnessed a defensive, defiant embrace of instructional strategies that turn back the clock, notably a focus on transmitting chunks of information to students - and doing so through direct instruction." But the evidence that supports this approach is slim and misleading. Take, for example, the oft-cited 'baseball study'. It has a tiny sample size, is narrowly focused, and assumes its own conclusion when it evaluates 'success'. And yet it's called "seminal". What really matters, says Kohn (and I agree) is that "getting kids to unpack or remember a specific text is a very different goal from helping them to become 'successful independent readers' over time. Moreover, knowing more stuff has a very limited role to play in helping students to read more proficiently, or think more clearly, or solve problems better."

While most people probably would not endorse Shopfy's approach of using a top-down edict to demand that everyone use AI, the example is nonetheless useful as a wake-up call, and that's how Carlo Iacono treats it here. "One key insight from Shopify's approach is the focus on human-AI collaboration as a source of innovation. Yes, there's an efficiency drive (doing more with less), but there's also a strong message of agency – empowering each individual to achieve more with AI, to level up their work in ways previously impossible." I'm not going to disagree with that. AI is a force-multiplier.

This is a study of what are obviously early adopters of Anthropic's AI by students in the U.S., so it's a glimpse, and not an authoritative sample, as the authors readily acknowledge. There are some interesting observations, for example, the uses AI is put to by students in different disciplines; computing science students will ask the AI to create or analyze code, while social sciences students will ask the AI to explain theories. I wonder, though, whether the sample didn't capture a fair number of teachers, because the two types of 'collaborative' use seem suspiciously like teaching tasks: "teach programming fundamentals with Python examples" and "provide feedback and revision for student writing assignments".

The focus of this article is the one Nature article on inhibitory cells, but you can read about the full set 10 open access articles on the mouse connectome released today. It's a good choice of article to focus on, because most people don't think of inhibitory cells when they think of neural nets (though they feature prominently in some types of artificial neural network). Data from the ten papers are accessible in an interactive online resource called the MICrONS Explorer (pictured). "The result is a map with an exquisite level of detail that researchers can use to probe how structural features shape connectivity, uncover local rules of circuit organization and test long-standing hypotheses about functional motifs that are repeated throughout the cortex." For example, as often cited in these pages: "Neurons that have similar firing patterns in response to a visual stimulus tend to connect with one another."