everwas

a blog by Ian Kennedy

Tag: artificial intelligence

  • Sitemaps for AI

    Sitemaps for AI

    Last week, I was double-booked in conferences. Wednesday & Thursday I was in Philadelphia for the beginning of the Online News Association conference, a gathering of journalists who work with words online. Friday & Saturday, I was in Washington DC for WordCamp, a gathering of people who work with WordPress, the CMS software that powers many of the websites journalists use to publish their news online.

    Hopping between these two worlds, the editorial and technical, gave me a unique perspective of the change sweeping online media. Everyone agrees that AI Chat Bots, specifically generative AI from Large Language Models (LLMs), will have an enormous impact on what we read online. But, depending on who you’re talking to, it’s going to result in either the horror or something wonderful.

    It’s still very early but a long Amtrak train ride home gave me some time to project out where we’re headed and ponder what we might need to make it work in a way that both publishers and AI Chat Bot companies feel comfortable.

    Those that fear AI view it as something that will strip mine websites of their facts and process them into the bland, robotic responses that power AI Chat Bots. This characterization echoes the publishing industry’s initial reactions to Google search. In 2006, French and Belgium newspapers demanded to be removed from Google News only to come back begging for inclusion in 2011 after they experienced a precipitous drop in traffic.

    Are we seeing the same thing play out with AI? Isn’t an AI chat bot just the conversational form of the Google SERP? Microsoft Bing Chat and Google Bard are crawling the web for tidbits to power their conversational engine. Concern about Bing and Bard abstracting facts without sending users back to a publisher’s site exposes a flaw in the publishing business model where a website is compensated by readers looking for answers on a page in adorned with advertisements designed to distract and harvest attention.

    Dare Obasanjo on threads.net

    It’s time to upgrade this business model. Instead of asking people to browse a bunch of search links, AI Chat Bots bring information to the reader, aggregated, summarized in a conversational tone. To a certain extent, this is an evolution of what has been happening for years.

    Google Knowledge Graph

    When Google Knowledge Graph launched in 2012, many publishers felt the Knowledge Panel (as it came to be known) did not provide enough attribution. Sound familiar?

    If the reader no longer goes to the publisher’s site, they will end up spending time with the product providing the answers, not the source. Back then it was Google, today, it’s the AI Chat Bot.

    The AI Chat Bot is the latest step in a journey that was started a long time ago. Bringing answers into a conversational UI is just improving on user experience for those in search of quick answers to their question.

    Bing Chat AI

    This new conversational UI is under rapid development. I’m not even sure a conversational is where we’ll end up. Microsoft is leading the way with Bing Chat AI results sprinkled with attributions that give credit and links back to the source material. From what I can tell, Microsoft is also is paying for this attribution in an early experiment in what I would call “licensing of facts.” Google’s Bard is following Microsoft’s lead and is also starting to add attribution to its SGE results, something that was missing at launch. I’d be curious to know if they are paying publishers for these links.

    Microsoft is embedding Bing AI in not only into their Edge Browser but has also announced extensions for Chrome and Firefox. Bing Chat is also available as an Enterprise service as well as on their mobile app and Skype.

    The pressure is on and Google is responding in kind with their version of generative AI chat, SGE, which is running in Google Labs.

    If generative AI is the next generation of search, I can think of a number of things that are needed to build a relationship between the publisher and AI vendor that is transparent, trustworthy and thus, sustainable. Allow me to riff a bit.

    Honor Robots.txt
    Open AI already announced that they would honor robots.txt and not crawl sites that declare themselves off limits. This is now extended to optimize which sections of your site you want to make available to the AI Chat Bots. The New York Times, CNN, and others are already adopting this method to control what they make available.

    This is a step in the right direction as it builds trust but more granular control over what is made available for the crawl is necessary. Within a restaurant review, maybe the address & phone number will be valued one way while the reviewer’s opinion valued another way.

    Sitemaps for AI
    A sitemap is a file that instructs a web crawler where to look for new pages. A sitemap for AI could be an intentional declaration by a site owner of what specific facts and information you want to make available and what link you want to serve up for the attribution. Addresses can be fielded and formatted one way, quotes another way so that they travel along with the name of the person quoted.

    Ads.txt was developed to make programmatic advertising more transparent. What I’m thinking of is something in between a sitemaps.xml and ads.txt, a lightweight, machine-readable way for publishers to declare what they make available to the Chat Bot crawlers.

    Real time Fact Exchange
    The technology that enables the real time auction for ad impressions on sites in milliseconds is some the most impressive technology developed for the internet in the past couple of decades. The incredible revenue machines of the ad industry have fueled the advancements in this technology.

    It’s time for a similar exchange for the facts which will be the new commodity. When looking for answers via a chat bot that has access to everything, maybe the deciding factor is the quality of the information or the party that is making it available. If every fact is distinct in the aforementioned Sitemap for AI, why not also attach a value to that fact that can inform the AI chat bot which information it can afford to share. If it’s a high value reader then more expensive information from higher quality sites might be presented. We are already headed down this path as both search results and social media links that go to paywalled sites attempt to capture subscriber budget.

    Is it finally time to create a marketplace of micro-transactions brokered by the Chat Bot UI? Instead of subscribing to a bunch of subscription sites, maybe the AI is where “pay” for tidbits of information with either advertising or payment tiers and that revenue is shared by the Chat Bot companies with the companies providing the information?

    In order for the Chat Bot AI ecosystem to grow, the publishers need to be fairly compensated and the Chat Bot vendors need a marketplace for the content they need to provide a quality experiences. Maybe the Real Time Fact Exchange is a far-fetched but I would have never thought the simple banner ad would have evolved into the complex ecosystem we have today.

    Further Reading:
  • ChatGPT Fun

    ChatGPT Fun

    I went to today’s State of the Word presentation in person (woohoo, so nice to attend in-person events once again!) and one question was about the dreaded “build-to-launch gap” where you build a website for a client and then it takes them months to post anything into the beautiful container you built for them.

    There was some suggestion of using ChatGPT to automatically generate content to fill the gaps. While it’s debatable that a boring prompt will lead to boring text, there is some use in using tools such as ChatGTP to automatically write website copy, the output of prompt with just a little bit of flair can be quite inspirations.

    While sitting there listening to the question and answer, I fired up ChatGPT and had the AI write something before my eyes. I unfortunately have only a screenshot of the first attempt which I showed to the amazement of the person sitting next to me as we saw the thing get typed up in real time.

    Just to try and re-create the scene so you could see how quickly it worked, I did it two more times. The first time it changed gears with the same prompt and wrote a sonnet about why you need to change your oil.

    The Prompt: Write a blog post about how important it is to change your oil every 3000 miles in the style of William Shakespeare

    Lend me your ears, fair readers, for I come bearing news of utmost importance. ‘Tis a tale of lubrication and maintenance, of the crucial role played by a humble substance in the wellbeing of your trusty steed.

    I speak, of course, of oil – that most vital of fluids, without which the inner workings of your car would grind to a halt and your journeys would come to a swift and unfortunate end. And yet, despite its importance, it is a substance too often ignored and neglected.

    Forsooth, I implore you – do not make the mistake of believing that oil needs changing only when your car begins to show signs of distress. Nay, ’tis a task that must be performed regularly, with the utmost diligence and care. The recommended interval, you ask? Why, every 3000 miles, without fail.

    Some may claim that this is an unnecessary expense, or that modern cars require less frequent oil changes. But I say unto you – the cost of a simple oil change pales in comparison to the expense of repairing the damage caused by neglecting this vital task.

    Think of it thus: the oil in your car is like the lifeblood that courses through the veins of a living being. It keeps all of the moving parts functioning smoothly, preventing friction and wear. Over time, however, this oil becomes contaminated and degraded, losing its ability to do its job effectively.

    When this happens, the various components of your car are left to fend for themselves, rubbing and grinding against one another with no protective barrier. The result? Reduced performance, decreased fuel efficiency, and a heightened risk of costly and inconvenient breakdowns.

    But fear not, dear reader, for all is not lost. By changing your oil every 3000 miles, you can ensure that your car is always running at its best. It may seem like a small and insignificant task, but trust me when I say that it can make all the difference.

    So take heed, and do not neglect this most essential of maintenance tasks. Your car will thank you, and you can rest easy knowing that you have done all that you can to keep it running smoothly.

    Related: I live-blogged this year’s State of the Word on my old tumblr account just for giggles

  • Fun with GPT-3

    Fun with GPT-3

    In April, Steven Johnson wrote that he latest generation of an algorithm that uses deep learning to produce human-like text called Generative Pre-trained Transformer, version 3 (GPT-3) could “write original prose with mind-boggling fluency.” I was intrigued.

    After reading Simon Willison’s excellent tutorial, encouraging me to play with GPT-3 because it’s “absurdly good fun,” I decided to dive in last weekend and play around. It was ridiculously easy to get started.

    I asked GPT-3 to come up with a new tagline for SmartNews based on the following prompt:

    Write a tagline for SmartNews, a news application that creates empathy for its users by showing them news from many different points of view. The algorithm does most of the curation but the entire process is monitored and enhanced by a small team of experienced journalists.

    Get the news you need, without the noise

    Not bad.

    I veered epistemological and asked GPT-3 to prove it’s intelligence. The following are screenshots from our conversation in the console which they call a Playground, reminding you that this is not the voice of God or anything like that and responses here are nothing to take too seriously.

    I like how my browser’s spell check stumbled over some punctuation.

    Cheeky eh? OK, right back at ya.

    I later read that this phrase, “There is no one correct answer to this question.” is a common dodge in order to duck a question that it does not feel confident in answering.

    Later on, on a call with my parents, I and explained how GPT-3 worked and asked them if they wanted to ask the algorithm some questions. My father fired off the question below.

    While the answer was the quality of a Google search result, the follow up response was quite interesting and shows not only the ability to summarize what it’s learned but also the ability to package and apply those learnings in response.

    The next question was also from my father.

    The algo’s response to the “complicated language” question was interesting. I did not think about the number of speakers of a language as an input but that was certainly thought provoking. I would whole-heartedly agree that Finnish is devilishly complex.

    I threw out the domino pizza question as a variation of Google’s famous how many golfballs fit inside a 747 question.

    I was surprised with how far off the answer was to this relatively easy to figure out question. (I’ve heard GPT-3 is not so great at math).

    I moved on to politics. Out of curiosity, I put this one out there just to see if had a sense of news & politics. It was not clear if this is ordered by priority but if it is, it’s interesting to see the #1 response which certainly seems timely.

    Incidentally, I asked the same question last night and got a different set of responses then from my questions which were asked last weekend.

    Inflation is certainly in the news and Yosemite is on fire. I’m not aware of any big cyber-attack but there are quite a few stories out there about potential Amazon Prime Day hacks. Nuclear War is interesting – could it be in response to PSA just released by NYC? Is GPT-3 reading the news?

    I have an on-going debate with my father about which city is more congenial, Tokyo or New York (we also have an on-going debate about print v. online but that’s for another blog post).

    The dialog is below. My father choose the word “comfortable.” I’m not sure about an algorithm’s definition of “comfortable” and you’ll see how it wriggles out of answering directly.

    Which city do most people think is more comfortable, Tokyo or New York?

    There is no definitive answer to this question as everyone has different opinions. Some people may find Tokyo more comfortable because it is a clean and safe city, while others may find New York more comfortable because it is more exciting and diverse. Ultimately, it is up to the individual to decide which city is more comfortable for them.

    Why is Tokyo clean and safe?

    Tokyo is clean and safe because it has strict laws and regulations regarding littering and public safety. There are also many public services that help to keep the city clean and safe, such as regular trash collection and a robust police force.

    Why is New York exciting and diverse?

    There are many reasons why New York is an exciting and diverse city. It is a melting pot of cultures, with people from all over the world coming to live and work there. There is always something to do, whether it is exploring the city’s many museums and galleries, or taking in a Broadway show. New York is also home to some of the best restaurants in the world, so foodies will never be bored.

    My mom asked a question which kicked off the most interesting exchange of the night when I tried to pin GPT-3 down on specifics.

    I’ll definitely be playing around with this some more.

    Firing off questions and pondering responses from the great algorithm in the sky is a big upgrade from the usual family activities such as playing charades or hearts. I’m bringing along GPT-3 to my upcoming reunion for sure.

  • Algorithms don’t get humor

    Algorithms don’t get humor

    April Fool’s Day is always a busy day at SmartNews as our news discovery algorithm get overly excited with all the cool, interesting and unique news released that day. Here’s a running log of what we’ve had to gently remind the algorithm is actually just a joke.

    Hope you had a good April Fools.

  • If HAL-9000 was Alexa

  • AlphaZero Masters Chess in Just 24 Hours

    AlphaZero Masters Chess in Just 24 Hours

    DeepMind, the same outfit that built AlphaGo, the AI platform that learned Go through supervised study of the game and went on to famously beat the top ranked player Lee Sedol has built an algorithm that now plays chess.

    What is even more incredible about this new “AlphaZero” AI is that it learned how to play chess through unsupervised learning. Instead of teaching it chess by feeding in key games and tactics, the designers just taught it the rules and let the algorithm figure out the best moves all on its own, by playing itself.

    Because it no longer needed to wade through and analyze historical data and also because it developed it’s own approach which was ruthlessly efficient. When AlphaZero was applied to Go, it surpassed AlphaGo within 3 days. AlphaZero was beating the strongest chess computer programs within 24 hours.

    instead of a hybrid brute-force approach, which has been the core of chess engines today, it went in a completely different direction, opting for an extremely selective search that emulates how humans think.

    Chess News writes about the development after reading a scientific paper published about the research accomplishment.

    In the diagram above, we can see that in the early games, AlphaZero was quite enthusiastic about playing the French Defense, but after two hours (this so humiliating) began to play it less and less.

    Chess News goes on to write about the broader impact of this breakthrough and what this means for the future of a generalized AI that can learn on its own.

    So where does this leave chess, and what does it mean in general? This is a game-changer, a term that is so often used and abused, and there is no other way of describing it. Deep Blue was a breakthrough moment, but its result was thanks to highly specialized hardware whose purpose was to play chess, nothing else. If one had tried to make it play Go, for example, it would have never worked. This completely open-ended AI able to learn from the least amount of information and take this to levels hitherto never imagined is not a threat to ‘beat’ us at any number of activities, it is a promise to analyze problems such as disease, famine, and other problems in ways that might conceivably lead to genuine solutions.

    Meanwhile, researchers at the University of Rochester have figured out a way to inject information into a monkey’s brain.

  • Reuters Tracer combs Twitter for news

    According to internal research, Reuters determined that 10-20% of news broke first on Twitter.

    Reuters, the news agency that first scooped its rivals with the use of carrier pigeons, is seeing good results from an algorithm to sift through Twitter (over 12 million tweets/day,  2% of total volume) to search for signal in the noise. Reuters Tracer is the system summarized in MIT’s Technology Review, How Reuters’s Revolutionary AI System Gathers Global News

    The first step in the process is to siphon the Twitter data stream. Tracer examines about 12 million tweets a day, 2 percent of the total. Half of these are sampled at random; the other half come from a list of Twitter accounts curated by Reuters’s human journalists. They include the accounts of other news organizations, significant companies, influential individuals, and so on.

    The next stage is to determine when a news event has occurred. Tracer does this by assuming that an event has occurred if several people start talking about it at once. So it uses a clustering algorithm to find these conversations.

    Of course, these clusters include spam, advertisements, ordinary chat, and so on. Only some of them refer to newsworthy events.

    So the next stage is to classify and prioritize the events. Tracer uses a number of algorithms to do this. The first identifies the topic of the conversation. It then compares this with a database of topics that the Reuters team has gathered from tweets produced by 31 official news accounts, such as @CNN, @BBCBreaking, and @nytimes as well as news aggregators like @BreakingNews.

    At this stage, the algorithm also determines the location of the event using a database of cities and location-based keywords.

    Once a conversation or rumor is potentially identified as news, an important consideration is its veracity. To determine this, Tracer looks for the source by identifying the earliest tweet in the conversation that mentions the topic and any sites it points to. It then consults a database listing known producers of fake news, such as the National Report, or satirical news sites such as The Onion.

    Finally, the system writes a headline and summary and distributes the news throughout the Reuters organization.

    Three recent events and their corresponding Tracer’s and Reuters alerts.

    More details (and attached screenshots) sourced from the paper, Reuters Tracer: Toward Automated News ProductionUsing Large Scale Social Media Data

  • Facebook F8 2016

    Facebook F8 2016

    Many years ago when broadband internet was still emerging, I spent an afternoon with a colleague in the company cafeteria trying to imagine a world with unlimited bandwidth and storage.

    We imagined that distances would collapse when the location of data would no longer matter. Music and video would be instantly available and you could call up anything you wanted to hear or see and jump to any point in a pre-recorded piece. Video conferencing would allow teams to work together, regardless of location. You could build connectors between data and services and create new views and from that gain new insights.

    Om Malik once proposed that broadband would serve as the railroads of our time. In the same way that the rail system in Europe and the interstate highway in the US mobilized industry and allowed remote communities to enjoy the output of industrialized centers, ubiquitous broadband would deliver the benefits of unlimited knowledge and ubiquitous reach to everyone around the world.

    facebook-f8-1

    At Facebook’s F8 developer conference we heard details of several projects which combine to bring internet to everyone around the world including Aquila, a drone that flies at 60,000 feet to extend connectivity to remote regions and Terragraph and Project Aries teaching telecom companies how to improve connectivity in crowded urban areas.

    facebook-vr-demo

    We also learned about projects that are being built to explore what can be done with this increased connectivity. The screenshot above is from a Virtual Reality demo in which we saw two people in different locations share an experience in a 360 virtual world, taking a selfie and sharing that “photo” to Mike’s Facebook wall.

    While the demo above is fantastic and paints a picture of what a shared virtual space might look like, it requires significant hardware and bandwidth to make happen. As people at Facebook like to say, this journey is only 1% finished.

    Oculus Research’s Yaser Sheikh talk on Social Presence in Virtual Reality that came at the end of the Day 2 keynote (59 min. into the video above) really brought everything together. The reason Facebook needs better connectivity is because they do not want to stop at having two avatars playing around in a fixed image 360 photo.

    To create a rich interaction where emotion and empathy can take place, we need to see all the subtle nuances that are expressed in the twitch of lip or roll of the eye. This is the unwritten language that we all know or what the anthropologist Edward Sapir called, “an elaborate code.”

    elaborate-code

    There is something visceral about interacting with someone in a shared space.  Yaser talked about the experience of his children in Pittsburg never really knowing his parents in India. To his kids, their grandparents are just, “moving images trapped behind a computer screen.” That is not how to build a lifelong relationship. Social VR aims to enable living and growing connections that are not a struggle to maintain.

    capture-display-predict

    There are three challenges to gaining a computational understanding of Sapir’s elaborate code.

    1. Capture – we need the ultimate motion capture of the whole body without being intrusive and in real-time. CMU’s Panoptic Studio is the state of the art but is still much too intrusive.
    2. Display – we need to transmit signals and animate avatars convincingly. The eyes, mouth, and hair are particular challenges.
    3. Prediction – we need and understanding of, “the vocabulary, the syntax, the morphology, and synchrony of social behavior” in order to write algorithms that help buffer social behaviors to overcome network latency (we all know how disruptive a bad connection can be to a video conference).

    Facebook’s ambition is to reverse engineer this elaborate code. While digital video streams a live image captured by a camera, virtual reality will capture, store, and animate a digital representation of someone. Words spoken and gestures shown are broken apart and recombined.

    Successfully building a prediction algorithm which can convincingly deliver requires an algorithm to continually anticipate state of mind and intent of others. This is much more than transmission of a moving image via bits – this is approaching the storage of the digital representation of what makes someone human. Building a library of all the possible human emotions and how to depict them is the ultimate moonshot and an appropriate one for a social network whose goal is to connect everyone. Stage one is capture and my sly take on the new Messaging Bot initiative is that all the conversations that are taking place on that platform are just step one in a big data harvesting program.

    Come full circle, back to that company cafeteria and imagine with me what a world would be like when Sapir’s elaborate code is cracked. When a digital avatar can be successfully animated we face some interesting questions.

    What royalties do you pay when a movie studio uses the digital representation of George Clooney instead of the actor himself?

    Can you simulate a debate between a virtual Donald Trump and a virtual Abraham Lincoln? If so, is it fair game to write about it and quote what Lincoln said?

    After Mark Zuckerberg is gone, will his employees consult his virtual avatar for management decisions? Are his avatar’s decisions contractually binding?

    Will a digital representation of someone understand humor? Sarcasm? What about a parody of recent events? Will tears well up as it tries not to cry?

    The Black Mirror episode Be Right Back explores what our relationship might be to a digital avatar (in this case to lost loved one) and is well worth a look if you haven’t seen it. While advances in technology can make the barriers of distance and time melt away so that we can keep relationships thriving, we must remember that the virtual world can never replace the real one and that there can never be a substitute for a face-to-face conversation.

  • Teach a Robot to See

    Following up on yesterday’s post about teaching artificial intelligence manners and ethics, this morning I listened to the latest episode of The Truth which features a dramatization of a blind man using a new kind of AI that helps him “see” – I won’t spoil the twist for you but it’s well worth the 10-minute listen.