The preconceived notions people have about AI — and what they’re told before they use it — mold their experiences with these tools in ways researchers are beginning to unpack.

Why it matters: As AI seeps into medicine, news, politics, business and a host of other industries and services, human psychology gives the technology’s creators levers they can use to enhance users’ experiences — or manipulate them.

What they’re saying: “AI is only half of the human-AI interaction,” says Ruby Liu, a researcher at the MIT Media Lab.

• The technology’s developers “always think that the problem is optimizing AI to be better, faster, less hallucinations, fewer biases, better aligned,” says Pattie Maes, who directs the MIT Media Lab’s Fluid Interfaces Group.
• “But we have to see this whole problem as a human-plus-AI problem. The ultimate outcomes don’t just depend on the AI and the quality of the AI. It depends on how the human responds to the AI,” she says.

What’s new: A pair of studies published this week looked at how much a person’s expectations about AI impacted their likelihood to trust it and take its advice.

A strong placebo effect works to shape what people think of a particular AI tool, one study revealed.

• Participants who were about to interact with a mental health chatbot were told the bot was caring, was manipulative or was neither and had no motive.
• After using the chatbot, which is based on OpenAI’s generative AI model GPT-3, most people primed to believe the AI was caring said it was. Participants who’d been told the AI had no motives said it didn’t. But they were all interacting with the same chatbot.
• Only 24% of the participants who were told the AI was trying to manipulate them into buying its service said they perceived it as malicious.
• That may be a reflection of humans’ positivity bias and that they may “want to evaluate [the AI] for themselves,” says Pat Pataranutaporn, a researcher at the MIT Media Lab and co-author of the new study published this week in Nature Machine Learning.
• Participants who were told the chatbot was benevolent also said they perceived it to be more trustworthy, empathetic and effective than participants primed to believe it was neutral or manipulative.
• The AI placebo effect has been described before, including in a study in which people playing a word puzzle game rated it better when told AI was adjusting its difficulty level (it wasn’t — there wasn’t an AI involved).

The intrigue: It wasn’t just people’s perceptions that were affected by their expectations.

• Analyzing the words in conversations people had with the chatbot, the researchers found those who were told the AI was caring had increasingly positive conversations with the chatbot, whereas the interaction with the AI became more negative with people who’d been told it was trying to manipulate them.

For some tasks, AI is perceived to be more objective and trustworthy — a perception that may cause people to prefer an algorithm’s advice.

• In another study published this week in Scientific Reports, researchers found that preference can lead people to inherit an AI’s errors.
• Psychologists Lucía Vicente and Helena Matute from Deusto University in Bilbao, Spain, found that participants asked to perform a simulated medical diagnosis task with the help of an AI followed the AI’s advice, even when it was mistaken — and kept making those mistakes even after the AI was taken away.
• “It is going to be very important that humans working with AI have not only the knowledge of how AI works … but also the time to oppose the advice of the AI — and the motivation to do it,” Matute says.

Yes, but: Both studies looked at one-off interactions between people and AI, and it’s unclear whether using a system day in and day out will change the effect the researchers describe.

The big picture: How people are introduced to AI and how it is depicted in pop culture, marketed and branded can be powerful determiners of how AI is adopted and ultimately valued, researchers said.

• In previous work, the MIT Media Lab team showed that if someone has an “AI-generated virtual instructor” that looks like someone they admire, they are more motivated to learn and more likely to say the AI is a good teacher (even though their test scores didn’t necessarily improve).
• Meta last month announced it was launching AI characters played by celebrities — like tennis star Naomi Osaka as an “anime-obsessed Sailor Senshi in training” and Tom Brady as a “wisecracking sports debater who pulls no punches.”
• “There are just a lot of implications that come with the interface of an AI — how it’s portrayed, how it interacts with you, what it looks like, how it talks to you, what voice it has, what language it uses,” Maes says.

The placebo effect will likely be a “big challenge in the future,” says Thomas Kosch, who studies human-AI interaction at Humboldt University in Berlin. For example, someone might be more careless when they think an AI is helping them drive a car, he says. His own work also shows people take more risks when they think they are supported by an AI.

What to watch: The studies point to the possible power of priming people to have lower expectations of AI — but maybe only so far.

• A practical lesson is “we should err on the side of portraying these systems and talking about these systems as not completely correct or accurate … so that people come with an attitude of ‘I’m going to make up my own mind about this system,'” Maes says.

Source: Placebo effect shapes how we see AI

Ben Welsh has a running list of the news organizations blocking OpenAI crawlers:

In total, 532 of 1,147 news publishers surveyed by the homepages.news archive have instructed OpenAI, Google AI or the non-profit Common Crawl to stop scanning their sites, which amounts to 46.4% of the sample.

The three organizations systematically crawl web sites to gather the information that fuels generative chatbots like OpenAI’s ChatGPT and Google’s Bard. Publishers can request that their content be excluded by opting out via the robots.txt convention.

Source: News organizations blocking OpenAI

Which reduces the value of AIs. It used to be the web was open for all, with information you could use as you liked. News organisations often fail to see value in AI but are scared that their jobs will be taken by AIs instead of enhanced. So they try to wreck the AIs, a bit like saboteurs and luddites. A real impediment to growth

[…]

I completely understand why some authors are extremely upset about finding out that their works were used to train AI. It feels wrong. It feels exploitive. (I do not understand their lawsuits, because I think they’re very much confused about how copyright law works. )

But, to me, many of the complaints about this amount to a similar discussion to ones we’ve had in the past, regarding concerns about if works were released without copyright, what would happen if someone “bad” reused them. This sort of thought experiment is silly, because once a work is released and enters the messy real world, it’s entirely possible for things to happen that the original creator disagrees with or hates. Someone can interpret the work in ridiculous ways. Or it can inspire bad people to do bad things. Or any of a long list of other possibilities.

The original author has the right to speak up about the bad things, or to denounce the bad people, but the simple fact is that once you’ve released a work into the world, the original author no longer has control over how that work is used and interpreted by the world. Releasing a work into the world is an act of losing control over that work and what others can do in response to it. Or how or why others are inspired by it.

But, when it comes to the AI fights, many are insisting that they want to do exactly that around AI, and much of this came to a head recently when The Atlantic released a tool that allowed anyone to search to see which authors were included in the Books3 dataset (one of multiple collections of books that have been used to train AI). This lead to a lot of people (both authors and non-authors) screaming about the evils of AI, and about how wrong it was that such books were included.

But, again, that’s the nature of releasing a work to the public. People read it. Machines might also read it. And they might use what they learn in that work to do something else. And you might like that and you might not, but it’s not really your call.

That’s why I was happy to see Ian Bogost publish an article explaining why he’s happy that his books were found in Books3, saying what those two other authors I spoke to wouldn’t say publicly. Ian is getting screamed at all over social media for this article, with most of it apparently based on the title and not on the substance. But it’s worth reading.

Whether or not Meta’s behavior amounts to infringement is a matter for the courts to decide. Permission is a different matter. One of the facts (and pleasures) of authorship is that one’s work will be used in unpredictable ways. The philosopher Jacques Derrida liked to talk about “dissemination,” which I take to mean that, like a plant releasing its seed, an author separates from their published work. Their readers (or viewers, or listeners) not only can but must make sense of that work in different contexts. A retiree cracks a Haruki Murakami novel recommended by a grandchild. A high-school kid skims Shakespeare for a class. My mother’s tree trimmer reads my book on play at her suggestion. A lack of permission underlies all of these uses, as it underlies influence in general: When successful, art exceeds its creator’s plans.

But internet culture recasts permission as a moral right. Many authors are online, and they can tell you if and when you’re wrong about their work. Also online are swarms of fans who will evangelize their received ideas of what a book, a movie, or an album really means and snuff out the “wrong” accounts. The Books3 imbroglio reflects the same impulse to believe that some interpretations of a work are out of bounds.

Perhaps Meta is an unappealing reader. Perhaps chopping prose into tokens is not how I would like to be read. But then, who am I to say what my work is good for, how it might benefit someone—even a near-trillion-dollar company? To bemoan this one unexpected use for my writing is to undermine all of the other unexpected uses for it. Speaking as a writer, that makes me feel bad.

More importantly, Bogost notes that the entire point of Books3 originally was to make sure that AI wasn’t just controlled by corporate juggernauts:

The Books3 database was itself uploaded in resistance to the corporate juggernauts. The person who first posted the repository has described it as the only way for open-source, grassroots AI projects to compete with huge commercial enterprises. He was trying to return some control of the future to ordinary people, including book authors. In the meantime, Meta contends that the next generation of its AI model—which may or may not still include Books3 in its training data—is “free for research and commercial use,” a statement that demands scrutiny but also complicates this saga. So does the fact that hours after The Atlantic published a search tool for Books3, one writer distributed a link that allows you to access the feature without subscribing to this magazine. In other words: a free way for people to be outraged about people getting writers’ work for free.

I’m not sure what I make of all this, as a citizen of the future no less than as a book author. Theft is an original sin of the internet. Sometimes we call it piracy (when software is uploaded to USENET, or books to Books3); other times it’s seen as innovation (when Google processed and indexed the entire internet without permission) or even liberation. AI merely iterates this ambiguity. I’m having trouble drawing any novel or definitive conclusions about the Books3 story based on the day-old knowledge that some of my writing, along with trillions more chunks of words from, perhaps, Amazon reviews and Reddit grouses, have made their way into an AI training set.

I get that it feels bad that your works are being used in ways you disapprove of, but that is the nature of releasing something into the world. And the underlying point of the Books3 database is to spread access to information to everyone. And that’s a good thing that should be supported, in the nature of folks like Aaron Swartz.

It’s the same reason why, even as lots of news sites are proactively blocking AI scanning bots, I’m actually hoping that more of them will scan and use Techdirt’s words to do more and to be better. The more information shared, the more we can do with it, and that’s a good thing.

I understand the underlying concerns, but that’s just part of what happens when you release a work to the world. Part of releasing something into the world is coming to terms with the fact that you no longer own how people will read it or be inspired by it, or what lessons they will take from it.

 

Source: Publishing A Book Means No Longer Having Control Over How Others Feel About It, Or How They’re Inspired By It. And That Includes AI. | Techdirt

 

A severely paralysed woman has been able to speak through an avatar using technology that translated her brain signals into speech and facial expressions.

[…]

The latest technology uses tiny electrodes implanted on the surface of the brain to detect electrical activity in the part of the brain that controls speech and face movements. These signals are translated directly into a digital avatar’s speech and facial expressions including smiling, frowning or surprise.

[…]

The patient, a 47-year-old woman, Ann, has been severely paralysed since suffering a brainstem stroke more than 18 years ago. She cannot speak or type and normally communicates using movement-tracking technology that allows her to slowly select letters at up to 14 words a minute. She hopes the avatar technology could enable her to work as a counsellor in future.

The team implanted a paper-thin rectangle of 253 electrodes on to the surface of Ann’s brain over a region critical for speech. The electrodes intercepted the brain signals that, if not for the stroke, would have controlled muscles in her tongue, jaw, larynx and face.

After implantation, Ann worked with the team to train the system’s AI algorithm to detect her unique brain signals for various speech sounds by repeating different phrases repeatedly.

The computer learned 39 distinctive sounds and a Chat GPT-style language model was used to translate the signals into intelligible sentences. This was then used to control an avatar with a voice personalised to sound like Ann’s voice before the injury, based on a recording of her speaking at her wedding.

The technology was not perfect, decoding words incorrectly 28% of the time in a test run involving more than 500 phrases, and it generated brain-to-text at a rate of 78 words a minute, compared with the 110-150 words typically spoken in natural conversation.

[…]

Prof Nick Ramsey, a neuroscientist at the University of Utrecht in the Netherlands, who was not involved in the research, said: “This is quite a jump from previous results. We’re at a tipping point.”

A crucial next step is to create a wireless version of the BCI that could be implanted beneath the skull.

[…]

Source: Paralysed woman able to ‘speak’ through digital avatar in world first | Neuroscience | The Guardian

Scientists have trained a computer to analyze the brain activity of someone listening to music and, based only on those neuronal patterns, recreate the song. The research, published on Tuesday, produced a recognizable, if muffled version of Pink Floyd’s 1979 song, “Another Brick in the Wall (Part 1).” […] To collect the data for the study, the researchers recorded from the brains of 29 epilepsy patients at Albany Medical Center in New York State from 2009 to 2015. As part of their epilepsy treatment, the patients had a net of nail-like electrodes implanted in their brains. This created a rare opportunity for the neuroscientists to record from their brain activity while they listened to music. The team chose the Pink Floyd song partly because older patients liked it. “If they said, ‘I can’t listen to this garbage,'” then the data would have been terrible, Dr. Schalk said. Plus, the song features 41 seconds of lyrics and two-and-a-half minutes of moody instrumentals, a combination that was useful for teasing out how the brain processes words versus melody.

Robert Knight, a neuroscientist at the University of California, Berkeley, and the leader of the team, asked one of his postdoctoral fellows, Ludovic Bellier, to try to use the data set to reconstruct the music “because he was in a band,” Dr. Knight said. The lab had already done similar work reconstructing words. By analyzing data from every patient, Dr. Bellier identified what parts of the brain lit up during the song and what frequencies these areas were reacting to. Much like how the resolution of an image depends on its number of pixels, the quality of an audio recording depends on the number of frequencies it can represent. To legibly reconstruct “Another Brick in the Wall,” the researchers used 128 frequency bands. That meant training 128 computer models, which collectively brought the song into focus. The researchers then ran the output from four individual brains through the model. The resulting recreations were all recognizably the Pink Floyd song but had noticeable differences. Patient electrode placement probably explains most of the variance, the researchers said, but personal characteristics, like whether a person was a musician, also matter.

The data captured fine-grained patterns from individual clusters of brain cells. But the approach was also limited: Scientists could see brain activity only where doctors had placed electrodes to search for seizures. That’s part of why the recreated songs sound like they are being played underwater. […] The researchers also found a spot in the brain’s temporal lobe that reacted when volunteers heard the 16th notes of the song’s guitar groove. They proposed that this particular area might be involved in our perception of rhythm. The findings offer a first step toward creating more expressive devices to assist people who can’t speak. Over the past few years, scientists have made major breakthroughs in extracting words from the electrical signals produced by the brains of people with muscle paralysis when they attempt to speak.

Source: Scientists Recreate Pink Floyd Song By Reading Brain Signals of Listeners – Slashdot

On Tuesday, Snapchat’s My AI in-app chatbot posted its own Story to the app that appeared to be a photo of a wall and ceiling. It then stopped responding to users’ messages, which some Snapchat users found disconcerting. TechCrunch reports: Though the incident made for some great tweets (er, posts), we regret to inform you that My AI did not develop self-awareness and a desire to express itself through Snapchat Stories. Instead, the situation arose because of a technical outage, just as the bot explained. Snap confirmed the issue, which was quickly addressed last night, was just a glitch. (And My AI wasn’t snapping photos of your room, by the way). “My AI experienced a temporary outage that’s now resolved,” a spokesperson told TechCrunch.

However, the incident does raise the question as to whether or not Snap was considering adding new functionality to My AI that would allow the AI chatbot to post to Stories. Currently, the AI bot sends text messages and can even Snap you back with images — weird as they may be. But does it do Stories? Not yet, apparently. “At this time, My AI does not have Stories feature,” a Snap spokesperson told us, leaving us to wonder if that may be something Snap has in the works.

Source: Snapchat’s My AI Goes Rogue, Posts To Stories – Slashdot

I do understand why so many people, especially creative folks, are worried about AI and how it’s used. The future is quite unknown, and things are changing very rapidly, at a pace that can feel out of control. However, when concern and worry about new technologies and how they may impact things morphs into mob-inspiring fear, dumb things happen. I would much rather that when we look at new things, we take a more realistic approach to them, and look at ways we can keep the good parts of what they provide, while looking for ways to mitigate the downsides.

Hopefully without everyone going crazy in the meantime. Unfortunately, that’s not really the world we live in.

Last year, when everyone was focused on generative AI for images, we had Rob Sheridan on the podcast to talk about why it was important for creative people to figure out how to embrace the technology rather than fear it. The opening story of the recent NY Times profile of me was all about me in a group chat, trying to suggest to some very creative Hollywood folks how to embrace AI rather than simply raging against it. And I’ve already called out how folks rushing to copyright, thinking that will somehow “save” them from AI, are barking up the wrong tree.

But, in the meantime, the fear over AI is leading to some crazy and sometimes unfortunate outcomes. Benji Smith, who created what appears to be an absolutely amazing tool for writers, Shaxpir, also created what looked like an absolutely fascinating tool called Prosecraft, that had scanned and analyzed a whole bunch of books and would let you call up really useful data on books.

He created it years ago, based on an idea he had years earlier, trying to understand the length of various books (which he initially kept in a spreadsheet). As Smith himself describes in a blog post:

I heard a story on NPR about how Kurt Vonnegut invented an idea about the “shapes of stories” by counting happy and sad words. The University of Vermont “Computational Story Lab” published research papers about how this technique could show the major plot points and the “emotional story arc” of the Harry Potter novels (as well as many many other books).

So I tried it myself and found that I could plot a graph of the emotional ups and downs of any story. I added those new “sentiment analysis” tools to the prosecraft website too.

When I ran out of books on my own shelves, I looked to the internet for more text that I could analyze, and I used web crawlers to find more books. I wanted to be mindful of the diversity of different stories, so I tried to find books by authors of every race and gender, from every different cultural and political background, writing in every different genre and exploring all different kinds of themes. Fiction and nonfiction and philosophy and science and religion and culture and politics.

Somewhere out there on the internet, I thought to myself, there was a new author writing a horror or romance or fantasy novel, struggling for guidance about how long to write their stories, how to write more vivid prose, and how much “passive voice” was too much or too little.

I wanted to give those budding storytellers a suite of “lexicographic” tools that they could use, to compare their own writing with the writing of authors they admire. I’ve been working in the field of computational linguistics and machine learning for 20+ years, and I was always frustrated that the fancy tools were only accessible to big businesses and government spy agencies. I wanted to bring that magic to everyone.

Frankly, all of that sounds amazing. And amazingly useful. Even more amazing is that he built it, and it worked. It would produce useful analysis of books, such as this example from Alice’s Adventures in Wonderland:

And, it could also do further analysis like the following:

This is all quite interesting. It’s also the kind of thing that data scientists do on all kinds of work for useful purposes.

Smith built Prosecraft into Shaxpir, again, making it a more useful tool. But, on Monday, some authors on the internet found out about it and lost their shit, leading Smith to shut the whole project down.

There seems to be a lot of misunderstanding about all of this. Smith notes that he had researched the copyright issues and was sure he wasn’t violating anything, and he’s right. We’ve gone over this many times before. Scanning books is pretty clearly fair use. What you do with that later could violate copyright law, but I don’t see anything that Prosecraft did that comes anywhere even remotely close to violating copyright law.

But… some authors got pretty upset about all of it.

I’m still perplexed at what the complaint is here? You don’t need to “consent” for someone to analyze your book. You don’t need to “consent” to someone putting up statistics about their analysis of your book.

But, Zach’s tweet went viral with a bunch of folks ready to blow up anything that smacks of tech bro AI, and lots of authors started yelling at Smith.

The Gizmodo article has a ridiculously wrong “fair use” analysis, saying “Fair Use does not, by any stretch of the imagination, allow you to use an author’s entire copyrighted work without permission as a part of a data training program that feeds into your own ‘AI algorithm.’” Except… it almost certainly does? Again, we’ve gone through this with the Google Book scanning case, and the courts said that you can absolutely do that because it’s transformative.

It seems that what really tripped up people here was the “AI” part of it, and the fear that this was just another a VC funded “tech bro” exercise of building something to get rich by using the works of creatives. Except… none of that is accurate. As Smith explained in his blog post:

For what it’s worth, the prosecraft website has never generated any income. The Shaxpir desktop app is a labor of love, and during most of its lifetime, I’ve worked other jobs to pay the bills while trying to get the company off the ground and solve the technical challenges of scaling a startup with limited resources. We’ve never taken any VC money, and the whole company is a two-person operation just working our hardest to serve our small community of authors.

He also recognizes that the concerns about it being some “AI” thing are probably what upset people, but plenty of authors have found the tool super useful, and even added their own books:

I launched the prosecraft website in the summer of 2017, and I started showing it off to authors at writers conferences. The response was universally positive, and I incorporated the prosecraft analytic tools into the Shaxpir desktop application so that authors could privately run these analytics on their own works-in-progress (without ever sharing those analyses publicly, or even privately with us in our cloud).

I’ve spent thousands of hours working on this project, cleaning up and annotating text, organizing and tweaking things. A small handful of authors have even reached out to me, asking to have their books added to the website. I was grateful for their enthusiasm.

But in the meantime, “AI” became a thing.

And the arrival of AI on the scene has been tainted by early use-cases that allow anyone to create zero-effort impersonations of artists, cutting those creators out of their own creative process.

That’s not something I ever wanted to participate in.

Smith took the project down entirely because of that. He doesn’t want to get lumped in with other projects, and even though his project is almost certainly legal, he recognized that this was becoming an issue:

Today the community of authors has spoken out, and I’m listening. I care about you, and I hear your objections.

Your feelings are legitimate, and I hope you’ll accept my sincerest apologies. I care about stories. I care about publishing. I care about authors. I never meant to hurt anyone. I only hoped to make something that would be fun and useful and beautiful, for people like me out there struggling to tell their own stories.

I find all of this really unfortunate. Smith built something really cool, really amazing, that does not, in any way, infringe on anyone’s rights. I get the kneejerk reaction from some authors, who feared that this was some obnoxious project, but couldn’t they have taken 10 minutes to look at the details of what it was they were killing?

I know we live in an outrage era, where the immediate reaction is to turn the outrage meter up to 11. I’m certainly guilty of that at times myself. But this whole incident is just sad. It was an overreaction from the start, destroying what had been a clear labor of love and a useful project, through misleading and misguided attacks from authors.

Source: The Fear Of AI Just Killed A Very Useful Tool | Techdirt

IBM and NASA have put together and released Prithvi: an open source foundation AI model that may help scientists and other folks analyze satellite imagery.

The vision transformer model, released under an Apache 2 license, is relatively small at 100 million parameters, and was trained on a year’s worth of images collected by the US space boffins’ Harmonized Landsat Sentinel-2 (HLS) program. As well as the main model, three variants of Prithvi are available, fine-tuned for identifying flooding; wildfire burn scars; and crops and other land use.

Essentially, it works like this: you feed one of the models an overhead satellite photo, and it labels areas in the snap it understands. For example, the variant fine-tuned for crops can point out where there’s probably water, forests, corn fields, cotton fields, developed land, wetlands, and so on.

This collection, we imagine, would be useful for, say, automating the study of changes to land over time – such as tracking erosion from flooding, or how drought and wildfires have hit a region. Big Blue and NASA aren’t the first to do this with machine learning: there are plenty of previous efforts we could cite.

A demo of the crop-classifying Prithvi model can be found here. Provide your own satellite imagery or use one of the examples at the bottom of the page. Click Submit to run the model live.

“We believe that foundation models have the potential to change the way observational data is analyzed and help us to better understand our planet,” Kevin Murphy, chief science data officer at NASA, said in a statement. “And by open sourcing such models and making them available to the world, we hope to multiply their impact.”

Developers can download the models from Hugging Face here.

There are other online demos of Prithvi, such as this one for the variant fine-tuned for bodies of water; this one for detecting wildfire scars; and this one that shows off the model’s ability to reconstruct partially photographed areas.

[…]

Source: IBM and NASA open source satellite-image-labeling AI model • The Register