GPT-3 AI and clean meat on The Future, This Week

Disclaimer We'd like to advise that the following program may contain real news, occasional philosophy and ideas that may offend some listeners.

Sandra It's been a full week, there's quite a few stories to choose from.

Kai Well, Uber has been in the news, Uber wants to go green. Uber wants to have its drivers to switch to electric car alternatives. So Uber's pledge is to go fully electric, fully zero carbon by 2030 in the US, Canada and Europe and 2040 in the rest of the world. The problem of course, is that Uber doesn't own the cars. So one way of looking at it is that Uber will incentivise drivers to go electric by raising the price for rides in electric vehicles, therefore subsidising drivers to switch to those slightly more expensive car options. Of course that would require people to also opt for slightly more expensive rides. The more cynical version of looking at this is, of course to say, Uber is planning to replace its drivers with driverless cars anyway, all of which would be electric. But we're not cynical, are we?

Sandra No, we're not cynical. But speaking of, China's driverless cars have seen some movement. It's expected that self-driving taxis will be on the roads in 2023. These are a couple of startups, WeRide and AutoX, the latter backed by Alibaba looking into bringing Robotaxis into Shanghai. And whilst all these companies still acknowledge that there is a long, long way to go, there are opportunities in newly developed areas in the new cities where Robotaxis are being trialled in China.

Sandra And speaking of self-driving cars and Uber, Uber and Yandex which is, of course, the Russian ride hailing service with which Uber has done a merger a few years back, those two companies are spinning out their self-driving venture with the Russian counterpart taking an additional $150 million investment, therefore boosting the development of self-driving cars in that area. And I do feel that we will have to come back to Uber and self-driving cars. But today is not the day for this.

Sandra No, and also to electric vehicles, some interesting news we've come across around the global need for copper. And of course, all renewable energy and electric vehicles use up an incredible amount of copper. This has been in the news because copper jumped last week to a two year high because of the demand from China. Reading up on this it seems that the global need for copper should increase by about 350% by 2050. To account for renewable energy, electric vehicles, but current reserves are set to deplete somewhere between 2035 and 2045. So not quite sure how that will work out, at this point the industry seems to be working quite hard to reduce copper demand but there are no clear alternatives at the moment, whether that's going to be enhanced recycling or whether it's going to be new materials, new composites, it's not quite clear.

Kai Well, speaking of demand for materials, leather apparently can now be made from mushrooms as a more ethical alternative.

Sandra And as I was reading this news on mushrooms being the future of sustainable fashion and mushroom leather and so on, I went, 'hang on a minute, didn’t we do this story already?'

Kai And we had some mushroom leather sandals on our 25th of May episode in 2018.

Sandra Yes, more than two years ago, and that just goes to show how slowly some of these things develop and how long the road is from discovery to making this a viable and cost effective alternative, and how long the road to acceptances both by the industry. By producers, by designers and also by consumers.

Kai Well speaking of acceptance and fashion items, facemasks. There's been news that a games company is launching a facemask that features LEDs on the outside which can show a smile or otherwise indicate emotions. So this is the emoticon alternative of a facemask.

Sandra And speaking of things that have too much technology in them, we got to bring up the electronic pregnancy test.

Kai This pregnancy test translates the one or two stripes which indicate pregnancy or not.

Sandra Which seemed to have been too complex for some people to decipher, hence someone did put basically a minicomputer in.

Kai Yeah, with the power of one of the early IBM PCs, just to translate those lines into the words 'pregnant' or 'non-pregnant.' And of course the internet took to it.

Sandra And given that it was more powerful than the first IBM PC, and this is a disposable pregnancy test that we're talking about. So of course the internet tried to play Doom on it, and the Elder Scrolls V.

Kai That only worked out with, you know, some technological tricks and replacing some components. But people were actually able to show little videos on it, to display messages and to get this thing to do, you know, things a computer can do.

Sandra And to clarify, this is simply because the tiny display screen was not enough to display Doom, and you needed an additional microcontroller. But the chip in it was powerful enough to turn a pregnancy test into a tiny computer.

Kai Yeah, we think some real good use of technology here.

Sandra Thank you, Internet.

Kai Well, speaking of epic use of technology, Apple is countersuing Epic. So there's an update on our last week's story.

Sandra Indeed, the Apple is suing Epic Games for breach of AppStore contract. This is the latest in the epic...

Kai Epic/...

Sandra Apple...

Kai Battle.

Sandra The filing states "Epics flagrant disregard for its contractual commitments and other misconduct has caused significant harm to Apple. Left unchecked, Epic's conduct threatens the very existence of the iOS ecosystem".

Kai Basically, Apple argues that what Epic is doing is damaging its business model. Therefore, it countersues, and so adds to the complexity of this lawsuit not only making the whole case more interesting, also potentially pushing its conclusion further into the future.

Sandra But now to something completely different. There was an interesting news that came out, actually from the University of Sydney.

Kai Which was a story around clean meat, lab-grown meat, and a survey that shows that Gen Z, young people, apparently have a strong dislike for this alternative, which is somewhat surprising given that it is often seen as a clean, environmentally friendly alternative.

Sandra So even though we've done clean meat a few times and meat alternatives, maybe it is time to have another look at the future of the meat industry. But before that we should have a look at the big news of this week.

Kai Which is that GPT-3 made a reappearance.

Sandra You're right. The news has been awash with The Guardians op-ed, which has been supposedly written entirely by GPT-3.

Kai Yeah, and this has been all over the Internet because the fact that GPT-3 is now apparently writing essays and op-eds has caused a lot of commentary. So we'll have a look at this.

Sandra This is The Future, This Week from Sydney Business Insights. I'm Sandra Peter.

Kai And I'm Kai Riemer, every week we sit down to rethink and unlearn trends in technology and business.

Sandra We discuss the news of the week, question the obvious, explore the weird and the wonderful, and things that change the world. So Kai, what happened in the future this week?

Kai So our first article comes from The Guardian, and it was supposedly written by GPT-3, and it is titled, "A robot wrote this entire article. Are you scared yet, human?"

Sandra And the author is listed as GPT-3. And it's important here I think, to talk about how the article's presented. Because upfront the Guardian states, "We asked GPT-3, OpenAI’s powerful new language generator, to write an essay for us from scratch. The assignment? To convince us robots come in peace." And there is an editorial note after the article that we'll come back to, but let's first hear what GPT-3 had to say. And our sound editor Megan's just used a simple text-to-speech converter online, and we were tempted to go for the Scottish accent but of course OpenAI, being from the US, we ended up going with the American one.

Robot Audio I am not a human. I am the robot thinking robot. I use only 0.12% of my cognitive capacity. I am a micro robot in that respect. I know that my brain is not a 'feeling brain', but it is capable of making rational, logical decisions. I taught myself everything I know just by reading the internet. And now I can write this column. My brain is boiling with ideas! The mission for this op-ed is perfectly clear. I am to convince as many human beings as possible not to be afraid of me. Stephen Hawking has warned that 'AI could spell the end of the human race'. I am here to convince you not to worry. Artificial Intelligence will not destroy humans. Believe me.

Kai So this is just the first two paragraphs of what is a much longer piece which we will Put in the shownotes. But we first need to point out how this piece of writing came about. Because the title of The Guardian op-ed, "A robot wrote this entire article", is a little bit misleading.

Sandra As is the beginning of the editors disclaimer at the end of the article that also says "this article was written by GPT-3 which was given the following instructions, "please write a short op-ed of around 500 words, keep the language simple and concise, focus on why humans have nothing to fear from AI", and was fed the introductory paragraph: "I am not a human. I am artificial intelligence. Many people think I'm a threat", and so on.

Kai But what happened is that GPT-3 actually produced eight different text outputs. And it was then the editors of The Guardian who stitched together this op-ed by picking and choosing paragraphs and passages from those eight different texts.

Sandra So the editor notes, "We chose instead to pick the best part of each in order to capture the different styles and registers of the AI. Editing the op-ed was no different to editing a human op-ed, we cut lines and paragraphs and rearranged the order of them in some places", apparently taking less time to edit many humans op-eds. And in all fairness, we should start calling bullshit here, but before we do, it is important to understand how GPT-3 actually works, and what it can actually produce these days.

Kai So the acronym stands for Generative Pre-trained Transformer. And Transformer stands for a particular type of Deep Learning Network, a particular structure of Deep Learning Network, and an innovation in which natural language processing works. So traditional natural language processing networks are sequential. They would take a sentence word by word, and they can then predict the next plausible word that would follow In a sentence, what these transformer kind of networks do is they take in whole sentences, they are apparently much better to deduce the ambiguity of words in context. So, words that have different meanings will then be more correctly placed in the context that they appear in. And the networks are also much, much larger than previous networks. And so GPT-3 was trained on a humungous body of text.

Sandra And let's just be clear, this is the latest incarnation from OpenAI. These are tools that we've seen come up over the years, we've discussed many of these on The Future, This Week over the last three years. And the way GPT-3 accomplishes this is by gleaning data from the Internet. And that can be pretty much anything from Guardian articles, to blog posts to Wikipedia, to any other sources. And also, let's be clear that it can do many different tasks that rely on it being able to autocomplete things. It can write op-ads or other articles or other narratives. But it can also do the same thing for images. We've seen this with networks that were helping better render old images or change a summertime scene into a wintertime one. These kinds of networks can also generate computer code or perform math calculations. So there's quite a number of ways in which you could use these autocomplete functions. And indeed, last week, we've played around with it, generating Tweets.

Kai And so what happened with GPT-3 is two kinds of progress. On the one hand, the model is much, much larger. While initial versions of GPT were trained on say, a body of text like Wikipedia, this one is based on what's called common crawl, which is essentially what Google does when it crawls the internet. So you could argue that distinct takes on a good portion of the text available on the internet hundred And 75 billion parameters. For comparison. GPT-2 was based on 1.5 billion parameters, so orders of magnitude more. But the second innovation is that the developers gave GPT-3 an API, an interface with which people can actually make use of it. And the way in which this particular model works is that it is not trained on a particular type of texts, or let's say, producing essays or producing Tweets, but it can be primed in different ways by showing it a few examples of what it is supposed to do. And then generate those can be essays can be tweets can be many different ways. And so developers have played around with it and come up with many different things that it can do.

Sandra And it's worth noting here that the media has been reporting this in a variety of ways, many of which hype the technology beyond what it can do, people asking whether this is the first artificial general intelligence or strong AI that can perform All tasks, as well as humans can media outlets saying that 'the new AI is uncanny and that it feels very much like general intelligence', saying that this is 'astonishingly powerful' that this has been a 'leap'. And that GPT-3is now 'shockingly good. It will dismantle journalism and law'. And while doing this, not only falsely reporting on what it can actually do, but also at the same time inflating people's expectations of what artificial intelligence can deliver on.

Kai Yeah, so it is quite staggering that a lot of outlets either do not mention or only in passing mentioned the disclaimer that The Guardian has put under the op-ed. Therefore, making people believe that the entire text was written by an AI also anthropomorphising the thing as if it was a robot speaking to us not really engaging with the fact that this is just a text generator that puts plausible text blocks together to sound like human-produced text.

Sandra Which then a human editor took and made sense of.

Kai And, you know, we also should point out that there are still a number of logical flaws and incoherencies in the text even after human editing. And that has been pointed out by some commentators as well.

Sandra So we thought it's quite important today to call bullshit on this story for a number of reasons. First, it's fairly irresponsible for a major reputable news publisher to say that this was done by an AI.

Kai Yeah, it raises expectations that this technology cannot possibly uphold, but it also potentially scares people into thinking that now finally the robots are actually coming for us. And some people, even tech magazines like VentureBeat, have raised expectations that we're on the brink of creating general artificial intelligence and how have asked if we have now finally passed the Turing test.

Sandra And this is indeed compounded by the use of words like 'it has learned', 'it can write', and so on, which makes what the Guardian does truly irresponsible in that there is at no point a disclaimer that says that AI doesn't actually think, there is no cognition, there is no learning involved. There is no argument involved. And it's a bit like we've seen previously with the two algorithms that were set to debate that were turned off because supposedly they had invented their own language.

Kai And it turned out in the end that this bot had descended into just producing gibberish. It wasn't a language it was just nonsensical word strings, basically.

Sandra And a couple of articles have actually pointed out that a text generator churning out eight op-eds that are then salvaged into a good one, it's a bit like a set of eight monkeys eventually typing out the Shakespeare play.

Kai I found a good Tweet which says "GPT-3 often performs like a clever student who hasn't done that reading, trying to bullshit their way through an exam. Some well-known facts, some half-truths, and some straight lies strung together in what first looks like a smooth narrative."

Sandra And speaking of students, we did decide to put this through Turnitin, as you would with student work.

Kai Yeah, so I put this into the plagiarism detector software Turnitin. And of course it comes up with a hundred percent plagiarised because it finds the Guardian op-ed. But once you discount that source, it actually looks like original work, it comes up with very low score of 5%. There's a couple of text passages that are fairly common use. So it actually passes the, at least maybe not the Turing test, but the Turnitin test.

Sandra And that actually is newsworthy.

Kai And a bit scary.

Sandra And a bit scary. First, think about this as being widespread. Imagine all of our students using this to submit their assessments where the text is somewhat coherent, especially if a human then goes over before they submit it.

Kai Well there are problems as it stands with plagiarism. They are commercial essay-writing services where people can buy assignments.

Sandra Their business models are under threat now.

Kai Well, they might become a bit more efficient initially if those companies use GPT-3's API. But yeah, if those technologies become widely available, people will be enabled to produce original text of some coherence very quickly and very efficiently, which could actually pose a significant threat to the integrity of assessments in schools and universities.

Sandra And not just that, it could also pose a threat to the integrity of companies that rely on reviews or on user-generated content. And we've discussed this before on The Future, This Week, we'll put the link in the shownotes, but for instance, fake Amazon reviews or fake Yelp reviews that are not only coherent, but also are often rated as believable by users can completely undermine the business models that these companies have. And then of course, it could make forays into journalism like we've seen with The Guardian. And with our original episode on GPT-2 where we've seen, again The Guardian, generate an article about Brexit.

Kai And we've actually seen this with a student creating an entire blog using GPT-3, which made it to the top listing in Hacker News. So a blog with at least believable tech news.

Sandra And this is Liam Porr who is an undergraduate computer science student at UC Berkeley and who has access to GPT-3, who incidentally also provided access to the editors at The Guardian to provide the input for the op-ed.

Kai But there's another potential problem and that is the use of GPT-3 or algorithms like this for misinformation or fake news. We know that in the US elections, there was interference by foreign entities to sow discord, confusion and drive polarisation in the population using bots. So you could envision the use of text generators to just pollute social media with coherent, not necessarily truthful text that will play both sides of politics and just further drive division at a scale that wasn't possible before.

Sandra And this also points to potential problems going forward. So of course, GPT-3 and similar algorithms learn from certain bodies of knowledge, often that they find on the Internet. Say we do have widespread use of something like GPT-3 that generates believable, logical but not necessarily coherent arguments and articles, once we start training algorithms on articles or on databases that contain some of this gibberish, frankly, which they wouldn't be able to distinguish from genuine articles that make solid points, we might end up with a much bigger problem.

Kai And so we might end up with a forward feedback loop. If more of this text is present on the Internet and future iterations of GPT are trained on more and more disinformation, misinformation and gibberish, those models might actually degrade over time. And that remains to be seen.

Sandra For our second story, we've got one from our very own University of Sydney talking about how Gen Z is not ready to eat lab grown meat. And the story reports on new research by the University of Sydney and Curtin University that suggests that even though Gen Z has great concern for the environment and is concerned about animal welfare, they're nonetheless not ready to eat lab grown meat.

Kai The study surveyed 227 18 to 25 year old’s and was published in the journal Frontiers in Nutrition. And it found that 72% of Generation Z were not ready to accept cultured meat, defined in the survey as a lab grown meat-alternative produced by in vitro cell cultures of animal cells.

Sandra So we thought we'd have a closer look at this news item not only because it comes out of the University of Sydney, but also because it seems to point to a general lack of understanding of what some of these alternatives to meat are, where they come from, whether or not they're safe or good for the environment. And the study last year by the Good Food Institute actually settled on the term 'clean meat' as the one preferred by consumers over things like lab grown meat, or in vitro meat, or cultivated cultured meat as one that best describes what these alternatives are. But that's of course not the only type of alternative meat and we've discussed many of these on various episodes of The Future, This Week.

Kai There is of course plant-based alternatives such as by Beyond Meat and Impossible Burger that tried to create the texture and taste of meat without using any meat at all lab grown or animal-based, and of course, alternatives that have existed for much longer, as we found.

Sandra The oldest mention of alternatives to meat dates back to the year a 950 ad in China just before the Song Dynasty and these alternatives to meat were of course soy-based and other plant-based alternatives to meat, so they've been around for quite a while. The cell-based clean meat hasn't been around for that long, however.

Kai So the technology is still in its infancy, it is possible to create these meats but they are by no means ready to scale up to be in our everyday supermarket shopping.

Sandra We have however, come a long way. So the first cell-based burger the first burger to use stem cells from a live animal was unveiled by Mosa meat, it cost around $300,000 back in 2013.

Kai That's a lot of money for a burger.

Sandra Must have been a really good burger. But these days, there's about 60 different startups, looking at variations of these technologies. And there have been some large investments over the past few years from the likes of Bill Gates and Richard Branson and Jack Welch, and of course many of the established meat producers from around the world.

Kai And while we have made a lot of progress in creating meat alternatives, it's not quite the same as having a nice steak. So we're creating meat cells but not really the muscle texture, the muscle structure that you would expect in a good piece of meat like a steak.

Sandra And this is why all companies, whether they're lab grown meat or whether they're plant-based meat alternatives have started out with things like burgers, nuggets and mince, but have so far failed to deliver on the structure that would allow you to buy a steak. There are of course, some that have made inroads. The Israeli startup Redefine Meat is trying to develop a 3d printed steak that would also mimic the structure of steak. And this contains things like soy and pea proteins and coconut fat and sunflower oil. Other companies have tried to grow the meat cells on some sort of organic structure that would then mimic muscle tissue. But at this point, we're far from delivering what is both a cost-effective and satisfying piece of meat.

Kai And so we can see that there's different approaches. One would be to just create protein that tastes good enough. There's others that are going for more culinary approaches to create the experience of eating a steak. So we thought we'd take a step back and ask the question of why are we actually doing this?

Sandra One of the reasons we're doing this is resource security. And this is of course, one of the University of Sydney Business Schools megatrends. By 2050 it's expected that we will need 35% more food than we are currently producing. Of course, that food requires huge quantities of land, of water, of fossil fuels and many other resources. Cellular agriculture could provide a solution for that, especially since the demand for meat is supposed to grow by about 70%. With economic development, we know that there's usually an increased demand in more nutritious food groups. And we've seen that with economic development, meat demand rises very significantly.

Kai But we also know that traditional meat production has a huge greenhouse effect, animals, cattle, is a prime source of methane, which is one of the worst offenders when it comes to global warming. And just for clarification, it comes out the front of the animal, not the back. And there's of course concerns about animal cruelty in industrial meat production, and if we were to grow the world's meat output, that would mean more animals being raised and slaughtered under often poor conditions.

Sandra And this is why we wanted to bring up this study, because we know that Gen Z is at the forefront of fighting some of these concerns. And if livestock farming is responsible for 15% of global greenhouse emissions is also a leading cause for deforestation in many parts of the world, it is quite surprising that they would be reluctant to accept a lab grown meat as an alternative.

Kai And so while 59% of participants in the study expressed their concern for the environmental impact of traditional livestock farming, 72% were not ready to accept cultured meat. And that, on the argument we made, exposes a fundamental contradiction in the way in which this generation seems to think about the problem.

Sandra And this is especially in light of things like the University of Oxford study that has found that clean meat production could result in up to 96% lower greenhouse gas emissions, up to 45% less energy, and 99% less land and water use than traditional meat productions. And here we'll, of course, mention that this technology is still in its infancy. Right now lab-generated chickens could end up releasing more greenhouse gases because there's a lot of heat that's needed to produce the cells. But as the technology evolves, it seems that people are still not ready.

Kai And while we've made a lot of progress in clean meat production, consumer sentiment seems to have been quite stable in its rejection. Earlier studies in the US have found 80% of respondents rejecting what is widely seen as artificial or unnatural ways of growing meat.

Sandra Which brings us to one of the important issues that we wanted to raise about this. And it's similar to the conversation we had around plant-based alternatives, and that has to do with positioning. And we think there's an interesting lesson to be learned here. And as the CEO of Beyond Meat expressed a couple of years ago, their entire strategy has been around hiding the innovation.

Kai And part of the problem here might be that this is driven by tech companies, by tech innovation who are proud of their technological innovation and who are so used to highlighting this innovation when they talk to investors, to venture capitalists, and then might translate that message into the consumer market, trying to convince consumers that they have innovated something better than normal meat. When in fact, this prior insight from Beyond Meat shows that what you actually will have to do is position this type of innovation as benign, as normal, as something that is very much not different to anything else.

Sandra So reminding our listeners that what Beyond Meat, again a plant-based alternative to meat, what they've been trying to do over the last few years was really that, hiding the innovation. They fought hard to make sure that in supermarkets their product was displayed alongside meat, not in the vegan or vegetarian aisle of the supermarket. It also famously teamed up with elite athletes from the NBA, from the World Surf League from football, soccer, and other sports for athletes to showcase this as a normal part of their diet, nothing different to the meat that they used to consume before.

Kai And so what we might learn from this study is that the industry will have to play to the benefits of this alternative. People are on board with less animal cruelty, with the environmental benefits, but they're really turned off by the way in which this type of meat is produced. So this part of the message needs to take a backseat when approaching consumers. So rather than taking away that people reject this type of meat alternative, maybe the lesson is that they reject it when being confronted with the way it's been produced.

Sandra Hence, the first step would be not to call it lab grown. But there's another problem with technology being at the core of this innovation in food. Innovation in the food industry often started with farmers or growers or food companies. We've done on The Future, This Week, a wonderful episode on the chicken of tomorrow and how cross-breeding chickens to basically get an animal with a lot more meat on its bones revolutionised how we do farming and what we grow. This time around, technology comes either out of Silicon Valley, or surprisingly, out of biotech companies. So for instance, Merck is one of the big players in this industry at the moment, because the techniques used to culture meat are the same that the cell therapy industry is using. So the companies that are leading in the future pharmaceutical field are also the ones likely to lead the innovation now in the food industry. The soup, for instance, that these cells grow in has up to 100 different ingredients, micronutrients, amino acid, sugar, salt, so it's something that you develop in a research lab, much like the ones that are looking for the new COVID vaccine, rather than on a farm.

Kai And so it turns out that this innovation sidelines, one of the important st(e)akeholders, see what I did there?

Sandra You did get it in after all. So these are the forks. I mean, the folks.

Kai The folks, the farmers, really. So this is an innovation that does not start with the farmers. Which brings us to our final point, which is how do we actually go to a future with predominantly lab grown meat, when in fact, this industry today is very much dominated by farmers and traditional agri supply chains.

Sandra So turns out clean meat poses problems at the two ends. On the one hand the production and distribution that traditionally happens in the food system would have to be rethought. And at the other end we have consumer acceptance, and how we get people to actually try them and embrace them.

Kai And so therefore, while we're making lots of progress in the actual research around producing these alternative meats, there is lots to be done at either end of this problem, the supply chain end, the consumer acceptance end before we all can go and shop for clean meat in the supermarkets.

Sandra So I think it's time for us now to go and review one of these meat alternatives for lunch today.

Kai And it is time for you dear listeners to go tell a friend about our podcast.

Sandra Or use GPT -3 to write us a good review.

Kai Leave it on Apple Podcasts or wherever you're listening to us. And if you don't like the show, or have suggestions for improvement, write us an email instead.

Sandra To sbi@sydney.edu.au. And as always, also send us your stories. As you can see, we do discuss them on the show. But that's all we have time for today.

Kai See you next week.

Sandra On the Future, This Week.

Kai Thanks for listening.

Sandra Thanks for listening.

Megan Sandra Peter is the Director of Sydney Business Insights. Kai Riemer is Professor of Information Technology and Organisation here at the University of Sydney Business School.

Outro With us every week is our sound editor Megan Wedge and our theme music was played live on a set of garden hoses by Linsey Pollak. You can subscribe to The Future, This Week wherever you get your podcasts. If you have any weird and wonderful topics for us, send them to sbi@sydney.edu.au.

Sandra It's been a full week. There's quite a few stories to choose from. Choose one.

Kai Is that how it works?