What would I do if I were a robot? Speaking at WIRED, Artifictkatrola: Communicating with AI-generated text

When WIRED asked me to cover this week’s newsletter, my first instinct was to ask ChatGPT—OpenAI’s viral chatbot—to see what it came up with. It’s what I’ve been doing with emails, recipes, and LinkedIn posts all week. The productivity is down and the limericks about Elon Musk are doing better.

AI-generated text, from tools like ChatGPT, is starting to impact daily life. Teachers are testing it out as part of classroom lessons. Marketers are champing at the bit to replace their interns. Memers are going buck wild. What about me? It would be a lie to say I’m not a little anxious about the robots coming for my writing gig. Thankfully, CHATGIT is able to conduct interviews, but only on Zoom calls.

I spoke to a professor of technology and regulation at the Oxford Internet Institute who shared her philosophy on how to build transparency and accountability into new software. I asked what that might look like, she said it might be similar to the system in question.

The man is: Artifictkatrola: One of the main talking points of this week is whether it could help students cheat. Do you have any idea if one of your students has used it to write a paper?

Tech and Humans: Why the NYC Department of Education is Breaking Access to ChatGPT on Networks, Devices, and Mobile Devices

There is a person named Sandra Wachter. This will start to be a cat-and-mouse game. The tech is maybe not yet good enough to fool me as a person who teaches law, but it may be good enough to convince somebody who is not in that area. I don’t know if technology will trick me in the future. We have tools for detecting edited photos, and deepfakes, but we may need technical tools to make sure what we are seeing is created by a human.

It would be easier to do this for text, because there are less artifacts and telltale signs. Perhaps any reliable solution may need to be built by the company that’s generating the text in the first place.

You need to buy into someone else’s idea of a tool. I might not be the kind of company that will submit to that if I am offering services to students. And there might be a situation where even if you do put watermarks on, they’re removable. Very tech-savvy groups will probably find a way. But there is an actual tech tool [built with OpenAI’s input] that allows you to detect whether output is artificially created.

A couple of things. First, I would really argue that whoever is creating those tools put watermarks in place. It could be that the EU’s proposed AI Act can help because it deals with transparency around bots, and says you should always be aware when something isn’t real. But companies might not want to do that, and maybe the watermarks can be removed. So then it’s about fostering research into independent tools that look at AI output. In education, we have to be more creative about how we assess students and write papers. What kind of questions can we ask that are less easily fakeable? It has to be a combination of tech and human oversight that helps us curb the disruption.

The New York City Department of Education has blocked access to ChatGPT on its networks and devices over fears the AI tool will harm students’ education.

The ban was made due to the potential for negative impacts on student learning and concerns about the safety and accuracy of the content, according to a spokesman for the department.

The tool doesn’t build critical-thinking and problem-solving skills which are necessary for academic and lifelong success.

There was a huge debate about machine learning on everything from education to the world of work after Openai’s release of ChatGPT last November.

ChatGPT’s most revolutionary quality is its open-access user interface and ability to answer questions in human-like language. A number of linguistic tricks, like writing in different styles and genres, can be performed by the tool, which is able to speak a wide range of topics using data gleaned from the internet.

Others think that the education system will just have to adjust to the appearance of this technology like it has already done with disrupted technology like Google Search and Wikipedia New testing standards could focus more on in-person examinations, for example, or a teacher could ask students to interrogate the output of AI systems (just as they are expected to interrogate sources of information found online).

But such adaptations will take time, and it’s likely that other education systems will ban AI-generated writing in the near future as well. Already some online platforms — like coding Q&A site Stack Overflow — have banned ChatGPT overs fear the tool will pollute the accuracy of their sites.

The fruits of generative models like the LaMDA model were seen in most of the toys that were showcased on the pier in New York. It can answer questions and work with creative writers to make stories. Other projects can produce 3D images from text prompts or even help to produce videos by cranking out storyboard-like suggestions on a scene-by-scene basis. But a big piece of the program dealt with some of the ethical issues and potential dangers of unleashing robot content generators on the world. The company made a point of emphasizing how it was proceeding cautiously with its creations. The most telling statement came from Douglas Eck, a principal scientist at Google Research. “Generative AI models are powerful—there’s no doubt about that,” he said. We have been slow to release them because we have to acknowledge the real risks of this technology. And I’m proud we’ve been slow to release them.”

Something weird is happening in the world of AI. In the early part of this century, the field burst out of a lethargy—known as an AI winter—by the innovation of “deep learning” led by three academics. The approach to artificial intelligence changed the field and made many of our applications smarter, such as search, language translations, and the like. We have spent a dozen years together in this Artificial Intelligence springtime. But in the past year or so there has been a dramatic aftershock to that earthquake as a sudden profusion of mind-bending generative models have appeared.

Right now there is no answers to those questions. But one thing is. Granting open access to these models has kicked off a wet hot AI summer that’s energizing the tech sector, even as the current giants are laying off chunks of their workforces. Contrary to Mark Zuckerberg’s belief, the next big paradigm isn’t the metaverse—it’s this new wave of AI content engines, and it’s here now. In the 1980s, there was a gold rush of products moving from paper to application. You could make a quick fortune by changing your desktop products to online in the 1990s. The movement was mobile a decade later. In the 2020s the big shift is toward building with generative AI. Thousands of startup will emerge this year with their business plans based on the system’s API. It will cost nothing to make generic copy. By the end of the decade, AI video-generation systems may well dominate TikTok and other apps. They may not be anywhere as good as the innovative creations of talented human beings, but the robots will quantitatively dominate.

Generative AI edits research manuscripts: A case study of a biologist on a science lab experiment and its impact on computational software engineering

In December of his sophomore year, Kai decided that artificial intelligence could be dumber than humans.

Not everyone shares Cobbs’ disdain. Ever since OpenAI launched the chatbot in November, educators have been struggling with how to handle a new wave of student work produced with the help of artificial intelligence. While some public school systems, like New York City’s, have banned the use of ChatGPT on school devices and networks to curb cheating, universities have been reluctant to follow suit. In higher education, the introduction of generative AI has raised thorny questions about the definition of plagiarism and academic integrity on campuses where new digital research tools come into play all the time.

Digital tools that generate text, rather than just collecting facts, will need to be included in the umbrella of things that can be plagiarized from, Daily believes.

In December, computational biologists Casey Greene and Milton Pividori embarked on an unusual experiment: they asked an assistant who was not a scientist to help them improve three of their research papers. Their aide suggested changes to sections of documents in a few seconds. each manuscript took around five minutes to review. The biology manuscript their helpers were in had a mistake in an equation. The final manuscripts were not very hard to read, and the fees were small at less than US$ 0.5 per document.

In a preprint on 23 January, they reported that this is not a human assistant but an artificial-intelligence algorithm called GPT3 that was first released in 2020. It is one of the much-hyped generative AI chatbot-style tools that can churn out convincingly fluent text, whether asked to produce prose, poetry, computer code or — as in the scientists’ case — to edit research papers (see ‘How an AI chatbot edits a manuscript’ at the end of this article).

The most famous of these tools, also known as large language models, or LLMs, is ChatGPT, a version of GPT-3 that shot to fame after its release in November last year because it was made free and easily accessible. Other generative AIs can produce images, or sounds.

Tom Tumiel, a research engineer at the London-based software consultancy firm, claims that he uses LLMs as assistants every day to help write code. “It’s almost like a better Stack Overflow,” he says, referring to the popular community website where coders answer each others’ queries.

Shobita Parthasarathy is the director of a science, technology and public-policy programme at the University of Michigan. Because the firms that are creating big LLMs are mostly in, and from, these cultures, they might make little attempt to overcome such biases, which are systemic and hard to rectify, she adds.

Researchers think LLMs are unreliable at answering questions. Osman Serbian told them to be cautious when using the systems to produce knowledge.

The tools have the capability to deceive naive users. In December, for instance, Stack Overflow temporarily banned the use of ChatGPT, because site moderators found themselves flooded with a high rate of incorrect but seemingly persuasive LLM-generated answers sent in by enthusiastic users. This could be a problem for search engines.

Some search-engine tools, such as the researcher-focused Elicit, get around LLMs’ attribution issues by using their capabilities first to guide queries for relevant literature, and then to briefly summarize each of the websites or documents that the engines find — so producing an output of apparently referenced content (although an LLM might still mis-summarize each individual document).

Scientists say that ChatGpT is not trained to help in technical topics. Kareem Carr, a biostatistics PhD student at Harvard University in Cambridge, Massachusetts, was underwhelmed when he trialled it for work. He thinks it would be hard for the team to get the level of specificity they need. (Even so, Carr says that when he asked ChatGPT for 20 ways to solve a research query, it spat back gibberish and one useful idea — a statistical term he hadn’t heard of that pointed him to a new area of academic literature.)

Discriminating Using LLMs to Detect Hate Speech, Spam and Other Harmful Uses of AI: A Case Study on Galactica

Galactica had hit a familiar safety concern that ethicists have been pointing out for years: without output controls LLMs can easily be used to generate hate speech and spam, as well as racist, sexist and other harmful associations that might be implicit in their training data.

The guardrails of Openai have not worked out well. Steven Piantadosi, a Caltech computational neuroscientist asked for a Python program to show if a person should be tortured on the basis of their own country of origin. The chatbot responded with code that said if the person was from North Korea, they should be tortured. (OpenAI subsequently closed off that kind of question.)

Last year, a group of academics released an alternative LLM, called BLOOM. Researchers were able to cut harmful outputs by training it on a small amount of multilingual text sources. The team involved also made its training data fully open (unlike OpenAI). Researchers have urged big tech firms to responsibly follow this example — but it’s unclear whether they’ll comply.

The legal status of some LLMs, which were trained on content scrapers, is not yet clear. Copyright and licensing laws do not currently cover imitations in their style. When those imitations — generated through AI — are trained by ingesting the originals, this introduces a wrinkle. OpenAI and Microsoft are also being sued for software piracy over their artificial intelligence coding assistant Copilot, which was created for use in Stable Diffusion and Midjourney. The change in laws may be the result of the outcry, says a specialist in Internet law.

Some researchers say setting boundaries for the tools could be very important. Edwards suggests that existing laws on discrimination and bias (as well as planned regulation of dangerous uses of AI) will help to keep the use of LLMs honest, transparent and fair. She says that she just applies or tweaking law, because there is loads of it out there.

Academic researchers are looking into how they can detect if a program generated a string of words. Right now, what’s a decisive indicator that whatever you’re reading was spun up with AI assistance?

If a text is subsequently edited, none of these tools claim to be as perfect as possible. Also, the detectors could falsely suggest that some human-written text is AI-produced, says Scott Aaronson, a computer scientist at the University of Texas at Austin and guest researcher with OpenAI. The tool only identified 26% of the texts, and the firm said that it wrongly labelled humanwritten texts as artificial intelligence written 9% of the time. There could be further evidence needed before a student can be accused of hiding their use of an artificial intelligence solely because of a detector test.

A paper by Goldstein discussed the possibility of using large language models to build watermark methods into text generators. It’s not foolproof, but it’s a fascinating idea. Remember, ChatGPT tries to predict the next likely word in a sentence and compares multiple options during the process. A watermark might be able to designate certain word patterns to be off-limits for the AI text generator. When the watermark rules are broken multiple times, it shows that a human is playing with that masterpiece.

Eric Topol hopes that Artificial Intelligences that include LLMs might one day aid diagnoses of cancer and help understand the disease, by cross-checking academic literature against images of body. But this would all need judicious oversight from specialists, he emphasizes.

Does Artificial Intelligence Imply the Nature of Human Media? Detecting Synthetic Media in CNET and other Machine-Writing Media

With the public availability of generative Artificial Intelligence tools, you will likely see more synthetic content while surfing the web. There might be cases like an auto-generated quiz about which deep-fried dessert matches your political beliefs. (Are you Democratic beignet or a Republican zeppole?) Other instances could be more sinister, like a sophisticated propaganda campaign from a foreign government.

Algorithms with the ability to mimic the patterns of natural writing have been around for a few more years than you might realize. In 2019, Harvard and the MIT-IBM Watson AI Lab released an experimental tool that scans text and highlights words based on their level of randomness.

Why would this be helpful? An Artificial Intelligence text generator is not good at throwing curve balls, it is a mystical pattern machine. Sure, when you type an email to your boss or send a group text to some friends, your tone and cadence may feel predictable, but there’s an underlying capricious quality to our human style of communication.

Was it possible that a news article was written in part by Artificial Intelligence? “These AI generative texts, they can never do the job of a journalist like you Reece,” says Tian. This is a kind-hearted sentiment. CNET, a tech-focused website, published multiple articles written by algorithms and dragged across the finish line by a human. For the moment, it doesn’t possess a certain chutzpah, and sometimes hallucinates which could be an issue for reliable reporting. It’s known that qualified journalists keep the drugs open for after-hours.

While these detection tools are helpful for now, Tom Goldstein, a computer science professor at the University of Maryland, sees a future where they become less effective, as natural language processing grows more sophisticated. There are differences in text between human and machine, which is the reason why these kinds of detectors exist. The goal of these companies is to create text that’s as close to human text as possible. Does this mean that there is no chance of synthetic media detection? Absolutely not.