Google opened its Cloud Data & AI Summit on Wednesday with a reminder that it was there at the start of the burst of enthusiasm for machine learning models – something it maybe feels has been lost now that so much attention is being paid to OpenAI’s GPT models and related Microsoft spinoffs.
“Foundation models have been years in the making, and Google has been at the very forefront of it,” said June Yang, VP of cloud AI and industry solutions at Google Cloud, citing the search megalith’s 2017 work on transformer technology, its 2018 BERT large language model, and DeepMind’s contribution to reinforcement learning.
Having established Google’s past contributions, Yang outlined a future role for Google Cloud that rests on chatbots and enterprise search.
Chatbots and AI-driven conversational interaction have been touted for years as the answer to … something. Rewind to 2018 when El Reg reported news of SAP testing an enterprise resource planning chatbot that chief commercial officer Franck Cohen hoped would help to automate about half of the German giant’s ERP system in three years. A year earlier, IBM promoted an AI-flavored career advice chatbot named Myca.
Next, step over the corpses of Facebook’s M AI assistant and Microsoft’s Cortana to the present enterprise market where cloud services promise businesses the chance to address their perennial interest in removing costly people from their balance sheets.
Yang argues that “foundational models” – large machine learning models that accept not only language but other inputs like images – represent a breakthrough in “the rapid democratization of AI and the creation of a new category of generative AI application, or what we call Gen apps.”
Generative AI refers to machine learning models that can “generate text, images, code, videos, audio, and more from simple natural language prompts.” Examples include ChatGPT, DALL-E, GitHub Copilot, or Google’s late entry into the fray, Bard.
For Yang, foundational models allow offerings like Google Cloud’s Vertex AI (a machine learning platform) and Gen App Builder (a no-code app builder for search and conversational interaction) to bring generative AI to enterprises.
“With Gen apps, organizations can pursue a whole set of new applications and customer experiences,” explained Yang. “Just as a shift from web application to mobile application made it possible for more people to access information and services from anywhere in the world, anytime, Gen apps are poised to enhance the way humans interact with technology.”
This is a more focused view than OpenAI’s ChatGPT plugin scheme which chains various services together through a common command prompt. Google Cloud’s use cases hew to the company’s long-standing corporate competence in search, and the conceit that conversational interaction works better in a customer service context than queries and clicks.
The shift from the desktop web to mobile cited by Yang is central to Google’s assumptions about the circumstances in which one might want to interact with a chatbot. One of the examples involved a scenario with a customer purchasing a bike using the mobile application of a fictitious vendor, Cymbal Bikes. The transaction took place as a chat conversation with a virtual assistant – the sort that might be built using Vertex AI and Gen App Builder.
Much of the proposed interaction could be accomplished just as well with a web form on a desktop device, where there’s enough screen real estate to place the interface elements necessary to make purchasing choices. But if your target audience really wants to engage on a mobile device, there’s an argument for capturing their preferences and purchase authorizations in a chat interaction.
Gartner VP and analyst Chirag Dekate told The Register in an email that chatbots have some advantages over static web pages.
“Generative AI augmented chatbots, that integrate domain knowledge, deliver more customized experience than a generic chatbot or a static website,” he wrote.
“For instance, being able to dynamically generate comparison tables, and remember interactions when consumers return to a website, are just some of many experiences that traditional chatbot and static websites cannot effectively deliver.”
Google Cloud will happily bill companies for creating and hosting their chatbots, but Dekate suggests the expense may be worth it. “The cost of engaging in Generative AI augmented chatbots that are easy to integrate and deploy could be easily made up from differentials experienced due to lost opportunity cost from static websites or legacy chatbots that frustrate potential clients,” he argued.
That may be, if chatbots don’t make you homicidal.
Another demonstration described how an analyst at a fictional investment firm, Cymbal Investments, might use a research app created by Gen App Builder to gather market data about the semiconductor industry.
“I start by asking in plain language which industries have been most impacted,” explained Lisa O’Malley, senior director of product management at Google Cloud.
“Now with one single prompt and interface I can see a variety of responses from both the internal and external data sources that my company has provided. Each entry has an AI-generated summary to help me quickly understand what’s important within it.”
Market research of this sort presently tends to involve search query competency as well as personal knowledge about reliable sources of information – not to mention individual data sifting and analysis skills. If business can replicate this process with a low-code prompt-driven system, it suggests that the individuals doing this job don’t bring much to the table and may end up losing their jobs to code.
Asked whether there’s any evidence that an AI-centric approach to enterprise search and customer interaction produces better results than otherwise, Dekate replied:
“Generative AI centered approaches that put enterprises in the driver’s seat by enabling them to isolate proprietary data from model data, and leverage domain knowledge to create delightful experiences can create transformative customer experiences. Highly personalized engagement experience, custom content that is sourced from domain knowledge and global knowledge are examples of how generative AI based approaches deliver a differentiated experience over their conventional counterparts.”
Dekate argued that Google has been at the center of data and AI since it became a thing, and has innovations to show for it. He added that Vertex AI and Gen App Builder offer possibilities that enterprises should explore.
“From a market cycle perspective, we are in a hyperactive exploration and development phase in Generative AI where continuous innovation and frequent updates are now the norm,” he wrote. “The ongoing Generative AI wars will be won by technology providers that successfully enable enterprises to customize applicability to their enterprise and industry context (with the necessary data, responsible and ethical AI guardrails) while insulating them from innovation risk.” ®