The Bard of Silicon Valley

Who was a Bard who changed the world: Not Shakespeare, but Google’s AI tool launched last week. We explore what it means for litigators and the broader world.

If you are coming here from LinkedIn, to continue where you left of click here.

AI based language models can be overwhelming or underwhelming, depending on your job security and whether you have actually tried using one. Let’s start with the good news for us humans. We do need to caveat this, as we are banking and finance experts, who are catching up on AI, and so there is scope for some of our technical descriptions to be wayward.

This nascent stage of AI tools isn’t much to write home about, at least in terms of immediate competition to mankind. AI, or more accurately the current release of machine learning tools, use a combination of language recognition and guided (supervised, unsupervised and semi-supervised) learning to attempt to converge to human behaviour.

A separate branch of AI is Deep Learning, which employs neural networks architecture to attempt to replicate the growth of more human intelligence. Traditional neural networks have a handful of layers, but deep learning tools may have over a hundred.

And so there we have it. This tool isn’t about replacing judgement. Rather it's language model can recognise text across the web (including in documents and filings), understand what they mean and link it to queries. Through a process akin to statistical regressions it will improve its understanding of user queries, and deliver more and more accurate results. This Google tool may make it's search tool obsolete, as it essentially sources and summarises large volumes of data. However, these are early days.

The best thing to do is to play with it. The link is below. As you will see some results are pretty inaccurate right now, however as users around the world use it, the programme is being 'trained' and so will improve.

Now for the bad news. If Goldman Sachs is right, or others in their camp, this jump alone will decimate the current workforce: 300 million white collar jobs are on the line, according to GS. The good news according to the GS report is that manual labour is safe. You will need to wait for the impending advances in robotics to address that issue.

Back to Google

Google has always been streets ahead of some of its major competitors. But it has been a little unproductive. The recent cost cutting drive laid bare what many suspected: that this extraordinarily successful company was not particularly ruthless in trimming non-productive assets and people and focusing on their winners.

For example Google’s suite of office tools is, at least in our view, far superior to Microsoft’s Office. As litigators (and lawyers generally) most of you can’t bear the idea of commenting on a Google Doc (an aspect where it is admittedly deficient). However, once you get beyond that, Google has an integrated Office Suite (Slides, Docs, Sheets, etc) that can all be weaved together with Scripts, a Javascript (coding language behind all websites) based tool that allows users to write code that automates all Google’s Apps.

Children at schools still learn on Microsoft’s Office, which is a tragedy given the power of opening these platforms to code: imagine it, you can programme your documents to grab data from spreadsheets, and even send emails from gmail, all in a bespoke fashion.

With such dominant products, Google should be converting young users around the world onto their platform, but it is Microsoft that is in all the schools (at least in the UK).

Microsoft’s purchase and release of OpenAI and products like ChatGPT has jolted Google into action. Hamstrung by ethical considerations, Google’s AI products to date sat within the confines of their vast R&D complex. Google X is an arm of the company whose sole purpose is the development of next generation products.

At Google’s IO event the announced the release of PaLM 2, their next generation large language model, that will power Bard (which will have coding capabilities to challenge GitHub for example - one of the worlds major code repositories, bought by Microsoft), it will have multi-linguality to reach 100 languages, Duet AI (a suite of AI tools for their office products), and a list of other major functions were previewed and discussed.

The speed of Google’s response to Microsoft’s purchase and release of OpenAI products, tells us, as many suspected, that within the labs of big tech are many technologies waiting to be released: Bard, google’s AI tools was one of them, and it is performing for the public now.

To trump Microsoft, Bard, was released with ready access to the internet. By contrast ChatGPT only had access to information that was on the internet prior to September 2021. Microsoft this week matched Google, by connecting ChatGPT to the internet. And so we get a sense of the unfolding race.

Tech is famously a space where the winners are often those who launch last, not first. It is a space where it pays to build a product, and test it comprehensively, before launching and growing quickly (which happens because the product is well developed).

That model does not appear to be true for machine learning. The first to launch gets a product that gets ‘trained’ by users, and so improves fast. The AI learning tool that gets the most user input will improve most quickly, gain more users, learn more and so a virtuous circle will result in the first to launch being the likely winner. Thus, Google, if it wished to play in this space, had no choice but to launch as soon as Microsoft had launched ChatGPT.

It is unclear to us how many more AI models are sitting in the wings, but given the need for these models to evolve by accessing user based ‘learning and training’ process, those that do not launch soon, risk having inferior products months and years down the line.

As Bard tells us when you sign up, it is an ‘experiment’, and will learn and improve over time. It certainly will.

Testing Bard

So we have played with Bard, as have others. It is far from perfect, but it is pretty cool. As we have argued before, if these are early versions, and given their modus operandi is to learn from users and improve, the potential for these tools is exciting and scary.

“It's really impressive. It's better than Chat GPT at this point from my experience on talk going through a number of things because it's actually connected live to the Internet it's connected live to search it can pull down real-time (ChatGPT caught up on this issue this week) data for you it can do real internet searches for you and just give you the results. It's extremely powerful I feel like it's the product that Google has been scared to do which is the

product that can truly disrupt search and they're doing it,” was how David Friedberg, Silicon Valley entrepreneur described it. We agree.

Why is it an improvement? Once again we will let Mr Friedberg explain.

“On the modelling side they're claiming a much larger more robust model it's linked to the to the internet. So it's linked to live search which you can't do with ChatGPT (which has no changed) which is based on a static training data set. So Bard can actually engage with dynamic content and a dynamic content generated across the web and then it's integrated

with a number of Google services (e.g. Youtube, that has a lot of data) that basically can take certain live feeds of data like flights and stock prices and so on it's free you don't need to be charged for for over usage you don't need to sign up and get access it's it's just available to anyone I mean I think this is the game changer everyone is hoping for,” Mr Friedberg explained.

There is a lot to unpack here. The first is the speed at at which competition took a paid model (ChatGPT was for subscribers last week) to a free model. The second is the rate at which content is being integrated into the AI learning model.

Google owns Youtube, and Youtube has vast amounts of content. With Speech-to-Text software also improving rapidly, all the content on Youtube can be condensed into a google query in no time.

So if you queried, ‘What does President Biden think about climate change?”, it can parse all the Youtube videos of speeches, as well as White House documents, and summarise and answer. Right now the Speech-to-Text on Youtube is deficient, and so it will not get Biden's views right (probably wrong in fact), but once these bridges are built, the potential is a game changer.

Our testing of Bard concurred with Mr Friedberg. Thus far it already performs the role of an untrained research assistant. Ask it if a major investor bought some shares, and it can check the internet, including public filings, and tell you. As it what profits or ebitda a said company made, and it can find the annual report, parse it, and let you know. It can quickly summarise facts, source lists of key data and so on.

And that is before you get to internal data. So, given banks already have access to internal emails and phone calls, it will likely change the world of internal monitoring and compliance monitoring. This of course makes the assumption that banks of today even exist in the future.

Further, the decision to give it the ability to form code and provide access to the office tools, is likely a response to the fact that most users who use wordprocessors (Microsoft Word) and spreadsheets (Excel or Google Sheets) don’t code, and so Google’s Javascript tools weren’t used. With this product that all changes. Ideally, a young lawyer may in the future be able to knit together functions across Google Docs, Sheets and Gmail. For example, take everyone’s billable hours, put it into a report and email it, all from code, and with no manual intervention, and put all that together without knowing how to code.

These are our initial thoughts as the race begins to hot up. The next decade is going to see these advances change the workplace. Technology gives smaller law firms that adopt it leverage to take on bigger players. It gives bigger players that adopt it become indomitable giants.

Governments and regulation

US Vice President Harris met with CEO’s of Alphabet, Anthropic Microsoft and OpenAI (now one firm), to underscore the responsibility of running these tools in a ‘trustworthy and ethical way with safeguards’.

Sam Altman (founder of OpenAI) suggested at a separate conference of the need for regulation on the level of the IAEA (International Atomic Energy Agency - the nuclear watchdog), for AI.

The Whitehouse has promised to issue guidance on AI oversight, which includes the National Science Foundation announcing USD 140 million to launch seven new National AI Research Institutes. FTC chair Lina Khan wrote an Op-Ed in the New Times about the need to regulate AI, watch for monopolies, consolidation, extortion, fraud and bias.