ChatGPT: A Chatty Generalized Productivity Tool ?
Marcus Weldon
Former CTO at Nokia and President of Bell Labs, exploring new vistas and providing Tech Strategy and Vision Consulting to the Good and the Great (of all types and sizes)
Over the past few weeks, I have been equal parts fascinated and surprised by the fervor with which the arrival of ChatGPT has been met, not only by technologists but also by the world at large.?What makes this remarkable to me is not only the speed at which this exuberance has developed, but also that it is not based on a revolutionary set of technologies, but rather an evolution of existing technologies (Convolutional Neural Networks, Generative Adversarial Networks, Transformers, Chatbots etc.), as pointed out by Yann LeCun in recent interviews.?What is remarkable about the emergence of these recent generative AI tools is that, typically, such technology evolutions show S-curve like adoption with a decade or more required to achieve widespread utilization.?So, what is going on with ChatGPT (and Midjourney, Dall-E 2 etc.)??How have they seemingly ‘crossed the chasm’ between early adopters and mainstream usage in a matter of weeks???And why is there almost a religious zeal to their arrival?
I have a working hypothesis to explain the fantastic fanaticism.?In short, I think they provoke and invoke a number of critical human reactions that go to the essence of humanity today, so there is a unique ‘resonance’ that has occurred that drives the rare reaction.?
Every good business owner will tell you that the core basis of any success has to be fulfilling, or solving for, an unmet ‘need’.?In the case of tools like ChatGPT, I actually think there are 4 essential human needs being met, which can be summarized as follows, in order of importance:
1.????The need for human communication
2.????The need to believe in something ‘more’
3.????The need for ‘assistance’
4.????The need for ‘new’
This may seem a bit overwrought but let me explain more.?Let’s start with the first two – the need for human communication and to believe in something more.?As Yuval Harari points out in his book ‘Sapiens’, the thing that separates us from our evolutionary precursors and cousins is the ability to communicate complex ideas to create a collective narrative, i.e., to form organized communities that can work towards a common purpose.?Furthermore, he points out that this ability results in the need to explain our existence and to account for phenomena that are beyond our control.?Specifically, he makes the argument that the rise of the earliest form of worship of something beyond ourselves coincided with the transition away from hunter-gatherer behavior towards settled agrarian communities, which were subject to significant exogenous effects that could wipe out their crops and consequently imperil their very existence.?And the only way to deal with such unpredictable catastrophe was to invoke the existence of a higher being responsible for these events that had to be appeased or lauded to prevent them from recurring.?So, there is a line that can be drawn between collective human communication and the foundation of belief systems that can induce extremely strong advocacy, or ‘fervor’.
It is clear that one of the unique characteristics of ChatGPT is the ability to sound human, and moreover like a rather intelligent human, with informed perspective on any issue.?In other words, ChatGPT is a great human communicator with a virtually limitless capacity for informed response, and the ability to learn from exchanges and mistakes – what more could one ask for???In this regard, it is clearly much more than the ‘dumb as a rock’ (in Satya Nadella’s words) agents such as Siri, Alexa, OK Google, etc., which merely use a human voice to serve up simple search results and programmed actions.?There are a wealth of studies that show how humans are so primed to communicate that if a machine or system attempts to communicate, even in a nonsense language, we are drawn to engage with it; we try to anthropomorphize anything and everything and are ecstatic when this produces something meaningful.?This is where chatGPT is essentially like human ‘catnip’.
But, I think the reaction to ChatGPT goes further than that. ?As suggested above, I think the recent fervor also stems from the belief in technology as a modern-day savior.?As we live in an increasingly secular age (at least in the Western world), we tend to worship at the altar of technology – technology that we believe has the promise to deliver us to a better, brighter future.?It is undoubtedly true that this belief in the power of technology to deliver us to some kind of nirvana has been significantly tempered by the questionable positions taken by the social media and web giants regarding their propagation of misinformation or hate-speech, as well as the proliferation of user profiling for financial gain, but these issues do not dissuade the tech faithful from their core beliefs and evangelism.?This neo-religious argument for the reaction to ChatGPT is very well made in a recent article by Jaron Lanier, in which he cautions against such worship of (technological) idols of our own creation.
Moving on from these rather philosophical and psychological observations to something more prosaic – the need for ‘assistance’ – it is a truism that humans invent machines to perform tasks for which people are not well-suited or even capable.?Indeed, this is the essence of the famous Moravec’s Paradox, which can be summarized as ‘human skills that appear effortless are difficult to reverse-engineer, but human skills that require effort may not be at all difficult to engineer’.?This is based on the straightforward observation that, on an evolutionary timescale, humans have optimized our abilities to exist and thrive in the physical world, but we have not (yet) evolved to be optimal in the digital-computational realm since we are just at the start of that journey. But even within the physical realm we have distinct limits in our individual and collective abilities. Therefore humans have created machines to be excellent assistants that offset and augment these limitations, e.g., on the physical side we have developed machines with mechanical advantage that allow us to manipulate massive objects or to move faster/go further; on the mental side, we have developed machines that allow us to compute and create at phenomenal speed.?And it is this latter category to which ChatGPT belongs, and is perhaps the best current example of such a human-augmenting system – one that has near-infinite access to information and can synthesize it into something closely resembling knowledge, anywhere, anytime: a hugely compelling value proposition.?
Now to the last ‘need’ ? the human need for ‘new’.?We are intrinsically exploratory in our make-up; we are wired to constantly search for and explore new ideas, new places, new tools, new constructs ? always seeking to improve something and increase our capabilities or efficiency at executing some task or process. ?
But what is the current ‘new’ need we have that is underserved ??With the unprecedented access to information and data at any time, we find ourselves increasingly overwhelmed by the task of curating that data into meaningful ‘knowledge’ or perspectives.?This ‘gap’ or need is at the heart of the vision for industrial and enterprise digitization (so-called ‘Industry 4.0’) – the need to enhance productivity using digital tools and assistants that curate the information we need for any particular task, exactly when and where we need it, and with the relevant context.?That task could be a physical task, in which case robotic assistants will be required to augment the human endeavor.?But in most cases, the task will be a creative task generating content in the form of images or text (or code), in which case a generative AI assistant would seem to be the ideal partner.?And this is precisely where the current swathe of generative assistants are beginning to show value – in the continuous creation and curation of content and information at scale and on demand.?Indeed, the potential (and pitfalls) of generative AI tools for enhancing productivity are covered in a very thought-provoking way in this recent article in the MIT Technology Review.
Now coming back to the beginning of this article, the conclusion can thus be drawn that the need for digital assistants such as ChatGPT is very real, but the fantastical fervor with which they have been received is only partly to do with the need they fill (and will increasingly fill), but also to do with the reaction we have to intelligent human communication and the desire we have to believe in something new that seems to have transcendent potential.?That is not to belittle the utility of ChatGPT and the plethora of related generative AI tools, but we do need to rationally assess their potential rather than imbue them with potential they do not (and will not) have.?My summary view of the rational utility of such tools is that they will be massively useful for the following types of tasks:
领英推荐
·??????Diverse example text and image generation
·??????Diverse information summarization
·??????Example code generation for diverse problems
·??????Creation of possible scenarios and solutions to stated problems
·??????Any other textual or image or data analysis or generation application for which absolute accuracy is not required
I particularly like this article that describes ChatGPT as providing a ‘blurry JPEG of the (2021 era) internet’ as one way of thinking about the real value.?I am also reminded of the distinction taught in school about the difference between precision and accuracy; I think generative AI is capable of statistical precision across a remarkable array of trained content, but is incapable of accuracy, or more correctly, incapable of even knowing whether it is accurate or not.?And increasing the size of the (already massive) training set is unlikely to result in higher accuracy as it simply doesn’t have the required reference framework to judge validity and therefore more input data will likely just result in the capture of more anomalous or extraneous information into the model. Don’t just take my word for this, Yann LeCun has recently argued that the existing generative technologies based on auto-regression (AR) methods are intrinsically flawed and get exponentially less accurate with problem size (length), so a new approach is needed that more accurately represents and predicts ‘reality’.?But that discussion is for another time.
In summary, I think ChatGPT should be thought of as an exceptionally good chatty tool to help with a large and general set of human queries and tasks, or to redefine the acronym, it is a Chatty Generalized Productivity Tool (ChattyGPT, if you like).?But, of course, we should ask ChatGPT whether this is fair characterization, and below is the exchange I had with it on this topic.?
Hard to argue with that summary.
Last, I also couldn’t resist seeing what Dall-E 2 could come up with when I asked it to ‘create an image showing ChatGPT being used as a productivity tool by a worker’ and here is what it created:
Clearly, there is much room for improvement on this front…?
Unlocking Potential Through Technology, Innovation, and Creative Collaboration
10 个月Looking back, I think the amount of hype is a function of humans mimicking other humans. ChatGPT has a broad general appeal, unlike most technology inventions it “seems” interesting to a large part of the human population. Hence the message gets amplified It doesn’t matter that the principles on which it is built aren’t ‘new’ - it’s that they haven’t been articulated like this before The world is currently in AI overshoot - NVidia GPUs are selling like crazy and a few hyper-scalers will have even more compute capacity than they didn’t really need. But what are the useful, value creating applications? AI is the new 5G - a part of me is jealous that it’s managing to attract most of the attention and investments right now. We’ll see what happens next I’ll be working on sustainable AI - cause it looks like we could use some
Technical Lead - Metaverse Platform Org
1 年In my experience, the need for 'assistance' is the primary factor driving the fervor for chatGPT. As noted, we live in the middle of an information explosion and there is a real need for tools to make sense of that information, use them properly, and even contribute. ChatGPT proved to be an amazing "information assistant" that can provide immediate value, unlike many other recent technologies like AR and VR. For this reason, I don't believe ChatGPT is hype but it is one of a series of technologies that will help humans overcome information problems.
Non-Profit Founder and Board Member
1 年Marcus Weldon what about asking GPT chat to get your voice summarised for us :-)
Global Strategic Alliances & Partnerships | Cloud & AI | Startup & Corporate Advisory
1 年I like your perspective on how ChatGPT fulfills human needs. Continuing on the human/machine symbiosis, I also enjoyed the recent analysis provided by Ben Thompson (at Stratechery) and the summary he makes of Jeff Hawkins' model of the human brain (which can be used to explain LLMs' "hallucinations"). You might also appreciate the references to Claude (Shannon): https://stratechery.com/2023/chatgpt-learns-computing/
Global Sales Lead @ Nokia - passionate about: building trust, delivering business value | specialized in digitalization: data center, 5G, IoT, edge cloud, OT/IT | Transformational leader
1 年And indeed, it’s almost addictive to play around with it - here the answer to the question to provide a smart comment on that blog posting: