Meet Bard – an early experiment by Google

Today, via email, I received an invitation to try Google’s new Bard Chatbot. After a limited, cautious demonstration in 2020, there’s now a limited public beta test. Because that protocol is the way to collect “valuable data” – at public scale. Sort of like a large clinical trial of a new medicine, eh [1].

---

Meet Bard, an early experiment by Google

Because you’re a Google One member, we’d like to offer you the opportunity to be among the first to sign up for the new Bard experience and provide feedback. Think of Bard as your creative and helpful collaborator, here to bring your ideas to life using generative AI.

[Button] Sign up to try Bard

You might ask Bard to outline a blog post about summer mocktail recipes, draft a packing list for a weekend fishing and camping trip, or help you understand if lightning can strike the same place twice.

We can’t wait to hear how people start using Bard, but we also know that large language models will not always get it right. Input from a wide range of experts and users will help Bard improve. Sign up to try Bard today.

---

Wired, among others, noted (below) this “beta” rollout. And reiterated some caveats [2]: “ChatGPT-style bots can also regurgitate biases or language found in the darker corners of their training data, …” And these bots tend to reflect back a user’s tone. What’s in a name: bard.

• Wired > “Google Rolls Out Its Bard Chatbot to Battle ChatGPT” by Will Knight (Mar 21, 2021) – The race to develop and commercialize the technology seems to be accelerating.

“Bard’s an early experiment, it’s not perfect, and it’s gonna get things wrong occasionally,” says Eli Collins, a vice president of research at Google working on Bard. … he did not specify how or exactly what restrictions Google has tried to place on the bot.

Collins … says one reason the company is launching Bard now, when the bot is far from perfect, is because of the valuable data generated when people interact with the system.

The bot will be accessible via its own web page and separate from Google’s regular search interface. It will offer three answers to each query—a design choice meant to impress upon users that Bard is generating answers on the fly and may sometimes make mistakes.

Google will also offer a recommended query for a conventional web search beneath each Bard response. And it will be possible for users to give feedback on its answers to help Google refine the bot by clicking a thumbs-up or thumbs-down, with the option to type in more detailed feedback.

---

Commentary

• Washington Post > “Say what, Bard? What Google’s new AI gets right, wrong and weird” by Geoffrey A. Fowler (March 21, 2023) – On its surface, the “prediction engine” Bard looks a lot like ChatGPT, the AI bot from research lab OpenAI.

[As a “responsible” rollout,] Google brings several advantages to the AI race. It has spent years developing its underlying chatbot technology called LaMDA, short for Language Model for Dialogue Applications.

Takeaways (article has examples for each)

• Bard can be smart, but still struggles [hallucinates”] with facts.
• Bard can show bias.
• Bard has guardrails, but can go off them.
• Bard works hard to not take sides.
• Bard displays a personality, but not humor.

• The Verge > “Google opens early access to its ChatGPT rival Bard — here are our first impressions” by James Vincent (Mar 21, 2023) – Google says the roll-out will be slow and has offered no date for full public access.

As expected, then, trying to extract factual information from Bard is hit-and-miss.

And how does Bard compare to its main rivals, ChatGPT and Bing? It’s certainly faster than either (though this may be simply because it currently has fewer users) and seems to have as potentially broad capabilities as these other systems. … But it also lacks Bing’s clearly labeled footnotes, which Google says only appear when it directly quotes a source like a news article and seemed generally more constrained in its answers.

• Forbes > “Bard Vs. ChatGPT: The Major Difference Between The AI Chat Tools, Explained” by Arianna Johnson (Mar 21, 2023) – Bard and ChatGPT have major differences, including the ability to remember past conversations and their knowledge base limits.

Stark differences

• Coding – not available in Bard as yet.
• Conversation Retention – Bard’s ability to retain context more limited.
• Responses – Bard’s LaMDA can draw responses from the internet (for latest responses) vs. ChatGPT’s dependence on a dated knowledge base [3].
• Language – Bard only (as yet) does English.
• Drafts – Bard presents several versions (drafts) for each response vs. ChatGPT’s single.

Here’s an interesting demo recap, an optimistic outlook with caveats.

• NT Times > Opinion > “Our New Promethean Moments” by Thomas L. Friedman (March 21, 2023) – This Promethean moment is not driven by a single invention, like a printing press or a steam engine, but rather by a technology super-cycle.

[Last week,] Craig Mundie, the former chief research and strategy officer for Microsoft [and a selected advanced tester], was giving me a demonstration of GPT-4, … Craig was preparing to brief the board of my wife’s museum, Planet Word, of which he is a member, about the effect ChatGPT will have on words, language and innovation.

GPT-4 exhibited an amazing level of originality during that demo – to summarize Planet Word and its mission in 400 (then 200) words (and then in other languages and forms).

Takeaways (unpacked in the article)

• An “Arthur C. Clarke” moment
• A “The Wizard of Oz” moment (with both a good witch and a bad witch)
• A Promethean moment [4] – a moment in history when “new tools, ways of thinking or energy sources” change everything.

We know the key Promethean eras of the last 600 years: the invention of the printing press, the scientific revolution, the agricultural revolution combined with the industrial revolution, the nuclear power revolution, personal computing and the internet and … now this moment.

… I call our Promethean era “The Age of Acceleration, Amplification and Democratization.”

The potential to use these tools to solve seemingly impossible problems … the way DeepMind, an A.I. lab owned by Google parent Alphabet, recently used its AlphaFold A.I. system to solve … the problem … known as protein folding. … ChatGPT is another such meta technology.

Like so many modern digital technologies based on software and chips, A.I is “dual use” — it can be a tool or a weapon.

The last time we invented a technology this powerful we created nuclear energy [developed & guardrail’d by (state) governments vs. (quasi-state) private companies] …

We are going to need to develop what I call “complex adaptive coalitions” — where business, government, social entrepreneurs, educators, competing superpowers and moral philosophers all come together to define how we get the best and cushion the worst of A.I.

Notes

[1] We’re cheap test subjects, eh. Self-randomized? No placebo? We’re writing the protocol on the fly. With a presumption to “rigorously monitor and assess outcomes.”

Costs for clinical trials can range into the billions of dollars per approved drug [or device].

Examples of clinical trial goals include assessing the safety and relative effectiveness of a medication or device [chatbot]:

• On a specific kind of patient [user]

• At varying dosages [screen times]

• For a new indication [change in behavior or emotion]

• Evaluation for improved efficacy in treating a condition as compared to the standard therapy for that condition [compared to standard search / social tech]

• Evaluation of the study drug or device relative to two or more already approved/common interventions for that condition [like particular search engines or social media]

A real example: Excerpt from an ad in LA Times 3-22-2023

AI Brings Clinical Trials to Cancer Patients at Warp Speed

MemorialCare Research Cuts Drive, Wait Time for Advanced Cancer Care Patients by More than Half with New TIME Trial

Tempus TIME Trial™ sequences a patient’s molecular data and clinical history and compares it against each trial’s structured inclusion and exclusion criteria through trial matching software and methods. This process of matching the patient to the clinical trial went from taking months to just 14 days. Tempus leverages its established relationships with hospitals across the country and pharmaceutical and biotechnology companies to increase patient access to the best fit clinical trials.

[2] As noted elsewhere, the chatbot is “a reasoning engine, not a fact database.”

[3] An update addressed this limitation.

• PC World > “ChatGPT’s new web-browsing power means it’s no longer stuck in 2021” by Mark Hachman, Senior Editor (Mar 23, 2023) – ChatGPT’s AI has one big problem: It was only programmed with information up to a few years ago. By adding web-browsing capabilities with its partner, Microsoft, OpenAI has brought ChatGPT up to date.

The update means that ChatGPT is expanding on several fronts, potentially eliminating a competitor’s advantage: Google’s Bard may be boring, but it knows what’s going on in the present, thanks to its integration with Google Search.

In one example, ChatGPT was asked about the latest Oscar winners. The app indicated that it was browsing specific sites to learn the information, even “clicking on” specific sources of information. Previously, ChatGPT was current up until about 2021, with certain exceptions.

[4] Wiki > Prometheus

In the Western classical tradition, Prometheus became a figure who represented human striving (particularly the quest for scientific knowledge) and the risk of overreaching or unintended consequences. In particular, he was regarded in the Romantic era as embodying the lone genius whose efforts to improve human existence could also result in tragedy: Mary Shelley, for instance, gave The Modern Prometheus as the subtitle to her novel Frankenstein (1818).