Artificial Intelligence of the Week: Microsoft puts an AI ad on your keyboard

Keeping up with a rapidly evolving industry like artificial intelligence is a daunting task. So until artificial intelligence can do that for you, here’s a handy roundup of the latest stories in machine learning, as well as notable studies and experiments we didn’t cover ourselves.

This week, in the field of artificial intelligence, Microsoft launched a new standard PC keyboard layout with “co-pilot” keys. You heard that right – going forward, Windows machines will have a dedicated key to launch Microsoft’s artificial intelligence assistant Copilot, replacing the proper Control key.

Some believe the move is intended to show Microsoft’s seriousness in investing in the race for consumer (and enterprise) artificial intelligence dominance. This is the first time Microsoft has changed the Windows keyboard layout in nearly 30 years; laptops and keyboards with copilot keys are expected to ship in late February.

But is this all a bluff? Windows users really? think Artificial Intelligence Shortcuts—Or Microsoft’s Artificial Intelligence Era Style?

Microsoft has undoubtedly demonstrated the infusion of “co-pilot” functionality into almost all of its products, old and new. Through flashy keynotes, slick demos and now AI keys, the company is highlighting its AI technology and betting on it to drive demand.

Demand is not a sure thing. But to be fair. Some vendors have successfully turned AI buzz into success. Take OpenAI, the maker of ChatGPT, for example. According to reports, the company’s annual revenue will reach $1.6 billion by the end of 2023. Generative art platform Midjourney is also apparently profitable and has yet to attract a penny of outside capital.

emphasize Some, although. Most vendors are weighed down by the cost of training and running cutting-edge AI models, and have to seek increasingly larger sums of money to stay afloat. For example, Anthropic is said to have raised $750 million in a funding round that would bring the total raised to more than $8 billion.

Microsoft, along with its chip partners AMD and Intel, hopes that AI processing will increasingly be moved from expensive data centers to local chips, commoditizing AI in the process — and probably rightly so. Intel’s new line of consumer chips includes custom-designed cores for running artificial intelligence. Additionally, new data center chips like Microsoft’s own could make model training cheaper than currently possible.

But there are no guarantees. The real test will be to see whether Windows users and enterprise customers, bombarded with Copilot ads, show interest in the technology and pay for it. If they don’t, it might not be long before Microsoft has to redesign the Windows keyboard again.

Here are some other noteworthy AI stories from the past few days:

• Copilot launches mobile version: In more Copilot news, Microsoft is quietly bringing the Copilot client to Android and iOS, as well as iPadOS.
• GPT store: OpenAI announces plans to launch GPT store, custom applications based on its text-generating artificial intelligence models (such as GPT-4)), within the next week. The GPT store was announced last year during DevDay, OpenAI’s first annual developer conference, but was delayed in December — almost certainly due to a leadership change that occurred in November after the initial announcement.
• OpenAI reduces supervision risk: In other OpenAI news, the startup hopes to channel most of its overseas business through Irish entities to reduce its regulatory risk in the EU. Natasha wrote that the move would undermine the ability of some EU privacy regulators to act unilaterally on concerns.
• Training the robot: Brian writes that Google’s DeepMind robotics team is exploring how to make robots better able to understand what we humans want from them. The team’s new system can manage fleets of robots working together and suggest tasks that the robotic hardware can accomplish.
• Intel New Company: Intel is breaking up Articul8 AI, a new platform company, is backed by Boca Raton, Florida-based asset manager and investor DigitalBridge. As an Intel spokesperson explained, Articul8’s platform “provides artificial intelligence capabilities that keep customer data, training and inference within the security confines of the enterprise” – which is ideal for highly regulated industries such as healthcare and financial services. An attractive prospect for customers.
• Dark fishing industry exposed: Satellite imagery and machine learning are providing new, more detailed insights into the shipping industry, specifically the number and activity of fishing and transport vessels at sea.It turns out there is Way New research published in the journal Nature by the Global Fisheries Watch team and several partner universities reveals this fact.
• AI-driven search: Perplexity AI, a platform that applies artificial intelligence to web searches, has raised $73.6 million in a funding round that values ​​the company at $520 million. Unlike traditional search engines, Perplexity provides a chatbot-like interface that allows users to ask questions in natural language (such as “Do we burn calories while sleeping?”, “What is the least visited country?”, etc.).
• Automatically written clinical records: More funding news, Paris-based startup Nabla Raised up to $24 million in funding.The company owns a Partner with Forever Medical GroupA unit of US healthcare giant Kaiser Permanente is developing an “artificial intelligence co-pilot” for doctors and other clinical staff that can automatically take notes and write medical reports.

More machine learning

You may remember various examples of interesting work from last year that involved making small changes to images that caused machine learning models to make errors, such as mistaking a picture of a dog for a picture of a car. They do this by adding “perturbations,” small changes to image pixels in patterns that only the model can perceive.or at least they idea Only the model can perceive it.

An experiment by GoogleDeepMind researchers showed that when artificial intelligence interfered with a picture of a flower to make it look more like a cat, people were more likely to describe the image as more like a cat, even though it definitely no longer looked like a cat . The same goes for other common objects like trucks and chairs.

Image Source: Google deep thinking

Why? how? The researchers didn’t really know, and the participants felt like they were just randomly chosen (in fact, the effects, while reliable, were barely above chance). It seems we are more perceptive than we thought, but this also has implications for security and other measures, as it shows that subliminal signals can indeed travel through images without anyone noticing.

This week, MIT conducted another interesting experiment involving human perception, which uses machine learning to help illuminate a specific language understanding system. Basically, simple sentences like “I walked to the beach” require almost no brain power to decode, while complex or confusing sentences like “In its aristocratic system, it influenced a dismal revolution” , resulting in greater and more widespread activation, as measured by functional magnetic resonance imaging.

The team compared activation readouts in humans reading various such sentences with the way the same sentences activated equivalent cortical areas in a large language model. They then made a second model that learned how the two activation patterns corresponded to each other. The model is able to predict whether new sentences will be cognitively taxing to humans. This may sound a bit mysterious, but it’s definitely a lot of fun, trust me.

Whether machine learning can mimic human cognition in more complex domains, such as interacting with computer interfaces, remains an open question. There is a lot of research out there, though, and it’s always worth a look. This week we introduced SeeAct, a system from researchers at Ohio State University that works by painstakingly basing an LL.M.’s explanation of possible actions on real-world examples.

Image Source: Ohio State University

Basically, you can ask a system like GPT-4V to create a reservation on a website, and it will understand what its task is and that it needs to click the “book” button, but it won’t really know how to do that. By improving the way it perceives interfaces with explicit labels and world knowledge, it can do better, even if it still only succeeds a fraction of the time. These agency models still have a long way to go, but regardless, expect a lot of big claims this year! I just heard some today.

Next, check out this interesting solution to a problem I didn’t know existed but makes a lot of sense. Autonomous ships are a promising area of ​​automation, but it can be difficult to keep them on track when the sea is angry. GPS and gyroscopes don’t solve the problem, and visibility can be poor – but more importantly, the systems that manage them aren’t overly complex. So if they don’t know any better, they can seriously miss their target or waste fuel taking a detour, which is a big problem if you’re running on battery power. I never thought about it!

Korea Maritime and Ocean University (another thing I learned today) proposed a more robust pathfinding model based on simulating ship motion in a computational fluid dynamics model. They propose that a better understanding of wave action and its impact on ship hulls and propulsion could greatly improve the efficiency and safety of autonomous maritime transport. It might even apply to human-guided vessels whose captains aren’t quite sure what the optimal angle of attack is for a given squall or wave pattern!

Finally, if you want a good recap of last year’s major advances in computer science (which overlap heavily with machine learning research in 2023), check out Quanta’s excellent review .