Findings from Silicon Valley part 5: How does Google work with AI

Visiting 谷歌 was interesting experience from multiple dimensions. First observation was that you don't always have to replace old ways with new ways. They can co-exist like those digital stairs and handwritten wall both telling their stories to people passing them. I think this is clever way to make impact and get the actual message noticed.

During presentations we had opportunity to drill in to the #HR side of this gigantic company. I think it is easy to understand that without well defined epectations (like those for leadership in above illustrated picture), it would be rather impossible to steer company as big Google into desired direction in cohesive way.

The #AI part of presentations was well structured and was nicely avoiding the hype without facts, which is often included in AI presentations. Different viewpoints were nicely captured like:

• Analysts gets intuitive access to complex data.
• Customer service can make online interactions more conversational.
• Creative parts of company can easily generate content (I don't know about this how creative the content will be at the end of the day, but it can be done fast)
• AI practitioners can customize foundation models.

Wide variety of different use cases related to these viewpoints were walked through as presented in picture above. They are quite generic as one could expect, but I think they still provide well covering illustration of the magnitude of the opportunities in this domain.

For me the last bullet was most appealing. If you can take large ready trained models, customize them for your own needs with the data you can control, you can add these generative capabilities natively into your ow internal machine learning platforms. This clean way to take AI into operational use will bring some tangible benefits for those organisations, who possess large amounts of data defining their own business environment.

Google's way of presenting their research work and the concrete results of it as family of models, is quite nicely showing the relationships between different phases.

• Foundation models are created for purposes depending on data: text, code, image, dialogue, etc. This underlaying work is often forgotten while end users are looking AI capabilities through the implementations where AI is embedded into chat or some other tool.
• Foundation models are used for powering solutions offered for consumers and enterprises. These use cases are general and thus can be utilized widely through different domains.
• AI professionals can then carry on building new more case specific solutions based on these domain specific models.

A concrete example of Google's BARD was given through generating summary of financial table of an example company. Not too fancy as it was just able to generate blank summary of presented numbers, but I think there are plenty of opportunities where to aim after this basic capability to handle input given like this.

Simple things like report generation are easily handled by AI. My take from this is that BI is one of the first areas which will be hit hard by AI starting to generate views to available data directly based on requests from business users.

After this insightful set of presentations (which I just briefly quoted here) about current state of Google's AI portfolio, we had opportunity to take bike ride to Google's new office building. Naturally with Google bikes.

The building looked fancy and modern with rainwater storages and solar panels in the roof. Like many other office spaces, it was basically deserted of people. Naturally we were only shown the common areas, so office-based innovation might happen hidden to our eyes. Yet I think freedom of choice can be quite important booster of innovation too and in general remote work seem to be popular also here. Different parts of the building were named based on animals and marked with huge statues of those. I think the "dinosaur district" was there on purpose.