Google is launching Bard, its version of Chat GPT, connected to search, and connected live to the internet. Sundar Pichai, CEO of Google and Alphabet, shared yesterday, “An important next step on our AI journey“. In discussing the release of Bard, Sundar said,

“We’ll combine external feedback with our own internal testing to make sure Bard’s responses meet a high bar for quality, safety and groundedness in real-world information.”

Following the link above led me to this next link:

“In addition to producing responses that humans judge as sensible, interesting, and specific to the context, dialog models should adhere to Responsible AI practices, and avoid making factual statements that are not supported by external information sources.”

I am quite intrigued by what principles Google is using to guide the design and use of Artificial Intelligence. You can go to the links for the expanded description, but here are Google’s Responsible AI practices:

1. Be socially beneficial.

2. Avoid creating or reinforcing unfair bias.

3. Be built and tested for safety.

4. Be accountable to people.

5. Incorporate privacy design principles.

6. Uphold high standards of scientific excellence.

7. Be made available for uses that accord with these principles.”

But these principles aren’t enough, they are the list of ‘good’ directions, and so there are also the ‘Thou Shalt Nots’ added below these principles:

I remember when Google used to share its motto “Don’t be evil”.

These principles remind me of the motto. The interesting vibe I get from the principles and the ‘Thou Shalt Not’ list of things the AI will not pursue is this:

‘How can we say we will try to be ethical without: a) mentioning ethics; and b) admitting this is an imperfect science without admitting that we are guaranteed to make mistakes along the way?’

Here is the most obvious statement that these Google principles and guidelines are all about ethics without using the word ethics:

1. Technologies that cause or are likely to cause overall harm. Where there is a material risk of harm, we will proceed only where we believe that the benefits substantially outweigh the risks, and will incorporate appropriate safety constraints.”

You can’t get to, “Where there is a material risk of harm, we will proceed only where we believe that the benefits substantially outweigh the risk“… Without talking aboutethics. Who is the ‘we’ in ‘we believe’? Who is deciding what benefits outweigh what risks? Who determines what is ‘substantial’ in the weighting of benefits versus risks? Going back to Principle 2, how is bias being determined or measured?

The cold hard reality is that the best Google, and Chat GPT, and all AI and predictive text models can do is, ‘Try to do less evil than good’ or maybe just, ‘Make it harder to do evil than good.’

The ethics will always trail the technological capabilities of the tool, and guiding principles are a method to catch wrongdoing, not prevent it. With respect to the list of things AI will not pursue, “As our experience in this space deepens, this list may evolve“… Is a way of saying, ‘We will learn of ways that this tool will be abused and then add to this list.

The best possible goals of designers of these AI technologies will be to do less evil than good… The big question is: How to do this ethically when it seems these companies are scared to talk directly about ethics?