Bard Updates and Googles Plans for AI
AI Leader Google gets whiplash, becomes Fast Follower
“AI is the most profound technology humanity will ever work on.” -Sundar Pichai
TL;DR - Google Bard isn’t as good as Bing Chat now, but it’s not standing still and will soon be more powerful. Google is moving deliberately to release new AI capabilities. So keep an eye on Bard as a potentially useful AI tool.
Bard Opens Up
Bard has opened up; it has been openly available to US and UK users since March 21 at https://bard.google.com. The PR for Google Bard hasn’t been kind since the demo that muffed a question and cost Google $100 billion in market value. There have been underwhelming comparisons to Bing Chat and chatGPT, and a Google employee quit and claimed Google copied chatGPT conversations in their AI development.
Google CEO Pichai has AI Thoughts
To respond to this, Sundar Pichai opened up about Google’s plans on the Hard Fork podcast. He explained that Google Bard rolled out with the LaMDA model and that they will soon upgrade Bard to utilize the more powerful PaLM model: “We have more capable models [PaLM] … You will see progress in the course of next week.”
The CEO shared a few interesting experiences with Bard. He asked Bard: "What do I do with my dad on an 80th birthday?" Pichai said Bard told him he should make a scrapbook for the event. “It's not that it's profound, but it says things and kind of sparks the imagination.” He also shared an experience with LaMDA pretending it was the planet Pluto:
"The way we primed it was 'Imagine you were Pluto as a planet,'" Pichai said. "And I remember playing around with my son at home talking to LaMDA back and forth. And there were a couple of conversations you really got deeply into it being Pluto. Because Pluto is far out in space, it became really lonely."
Other points Google CEO Sundar Pichai made about Bard and their AI strategy:
Bard will be getting better, quickly. “We will have coding capabilities in Bard very soon.” and “Search and AI will be integrated.”
On balancing innovation and safety, Google is “being very deliberate and responsible on releases.” He said “We are being careful with Bard.” At the same time, they set up green lane to approve AI products to knock down bureaucratic roadblocks.
They want the company to “move through this AI moment” in a responsible way. “The work on Privacy, safety and responsible AI is more important, our commitment is unwavering.”
“This has been a long journey for us” - referring to Google’s near 25-year journey in machine learning, deep learning and AI.
Regarding the “pause” letter, he believes there is “Merit to be concerned” about AI Safety, and “the spirit of the concerns are good to be out there.” At the same time, Google will not unilaterally stop development and he doesn’t see how such a pause could be implemented and enforced.
He disagrees with “being opinionated early on” about AI risks. He said “you have to worry about AI safety” even with considering getting to AGI. “It is clear to me that these systems will be very capable … they can cause dis-information at scale.”
He also mentioned how in the genetics field in the 1970s, leading researchers got together to establish a voluntary framework of rules. It would make a lot of sense to develop AI development principles that covers AI Safety, Alignment, Security, Privacy, Transparency and Ethics and get voluntary agreement to them, perhaps using existing AI guidance such as Asilomar Principles.
“AI is too important an area not to regulate, it is also too important an area not to regulate well.” He mentions also that you can leverage existing regulations to apply to AI, such as in healthcare and privacy regulations.
Regarding people seeing AI as sentient, such as the Google Engineer who felt that LaMDA was sentient: “You will see more like this,” he said. “This is the tip of the iceberg.”
To the question “When Bard gets good, how does the web survive?” He replied, “The traffic we send outside has only grown. … Google is always about helping you in ways that make sense to you. We will not be the be-all and end-all.”
Integrating Bard with Search
One would think that Google’s strength in search would make it natural for them to start with integration in search. Yet my own limited experience with Bard was disappointing on this front. Bard was able to inform me of specific flight information from a query, but then when I asked for the web links so I could actually book flights, it couldn’t help.
I suspect this will be fixed in short order, since the Google CEO says “Search and AI will be integrated.” However, Google also indicates they will be careful with how they proceed. They need to me. It’s not just about getting the product right, Google needs to be careful because it will be impacting their search business.
Whatever Google does, the Search Engine will evolve into the Answer Engine, and users will migrate elsewhere if Google doesn’t step and provide a top-notch solution.
Google’s Contributions to AI
Google has perhaps contributed more to the development of AI in recent years than any other company. Google has been using machine learning for the past 20 years to improve search results. Since 2012, Google used deep learning to solve problems in language translation, speech recognition, language understanding, image recognition and more, and in the process made many research contributions in the field of AI.
Google’s research on deep-learning language models led to their work on Transformers, which led to BERT, and then one of the first LLMs, T5, which paved the way for the LLMs we have today, including the LaMDA, a large language model behind today’s Bard.
They also developed image and video recognition to make sense of YouTube videos, help users find addresses in Google Maps, and applied image recognition to health-related problems, such as detecting diabetic retinopathy and cancerous cells.
Google's DeepMind team has also made many ground-breaking contributions, first conquering the game of Go with AlphaGo, and then in 2016 developing AlphaGo Zero, which learned to play Go without any human data, and went on to defeat AlphaGo by 5-0. In 2017, Google's DeepMind team developed AlphaFold, which predicted the structure of proteins with unprecedented accuracy, a major breakthrough in the field of biology.
Google’s technical strength and depth in AI is unparalleled, which is why we should not discount Google in the “AI race”. They are proceeding deliberately. They are often better at engineering than at developing and marketing great products. OpenAI and Microsoft stole mind-share with world-changing AI releases in recent months, but the Googleplex is striking back.