Mary Ann Allison

Governing AI series:  Part 2

This article is Part 2 of a series about the power of using multiple methods of regulating AI.   This is not, for example, about the content of a law or a policy recommendation from an AI developer industry group.  It is about what types of governance have a chance of working and why.

If you haven’t read Part 1 of this series, please do so here. This will then make much more sense.

You don’t have to be an AI developer, think-tank specialist, or legislator to read this.  Understanding how to regulate AI is important for everyone.

Whether you’re an AI developer, a think-tank specialist, a legislator, or a member of the general public, this is for you.  Understanding how to regulate AI is crucial for everyone.

The printing press was only the start.

In order to say, “it doesn’t say that,” we must:

• be able to read,
• have a book,
• have the skills to think effectively about what we’ve read, and
• be willing to act.

There’s a similar path in order to say “That’s a deep fake.”  But in this case, acting means developing the skills to recognize deep fakes and making the decision not to get sucked in.   We don’t get cheated and, if enough of us ignore fakes, the incentive to produce them is significantly reduced.

That’s a defensive strategy.  It is important.  Equally, or even more important, are positive activities.  They lend our weight to counterbalancing large institutions.

Both require actions—small and large—in the world.  The Industrial Revolution and the Protestant Reformation didn’t happen until people applied what they learned in their reading to the real world.  Until they started building machines and forming their own churches.

First, let’s understand the situation.

AI is rapidly increasing in power.  Many of us, and even some of AI’s key developers, are concerned about the future consequences.  When we feel fear, it is a signal that we think something is important. 

Unfortunately fear also often triggers our sympathetic nervous system and pushes us to leap into action when we really should be more thoughtful.  When reacting to an immediate physical threat, our fight-flight-freeze reactions may be helpful. 

They are not helpful when thinking about how to live in the world with AI.  It is important that we slow down so that we understand as clearly as possible what is actually happening before we take action.  We want our work to count.

Deep fakes have spooked us.

We see danger everywhere in the media. Many of us have lost faith in our ability to find accurate information or for any news source to represent reality without bias. We’re afraid that media company algorithms hijack our attention and promote discord. We may have lost faith too soon. We may have over generalized our response.

Our media landscape is extremely complex. Single cause and effect relationships are extremely rare, maybe non-existent.

Consider this example: During the 2016 US presidential election, many were concerned that extreme websites and videos polluted public opinion. However, recent research from social scientists shows that only 1% of Twitter users saw 80% of the dubious websites that were viewed during this time. Only a small percentage (6.3%) of YouTube users watched most (79.8%) of the extremist channels.

And, it was not, as often thought, that media platform algorithms promoted content with extreme views. In fact, less than 1% of YouTube’s algorithmic recommendations pushed people to extremist videos. Most recommendations were politically middle of the road. There is reason to believe that people arriving at extreme media sites are confirming their opinions, not developing new ones.

When we are planning to act, we will be more effective if we understand what is happening and use our resources where they will be most effective. However, research like this takes time.

What to do? Let’s understand that we are spooked. And that deliberate digital distortion does happen, can be harmful, and may be on the increase. Even if in fewer cases than we imagine.

When we feel safe, we relax. We are able to gather more information, to think before acting. I propose we increase our feelings of safety by taking action individually. Here are two tactics to start with—one defensive and one that builds positive momentum. The AI landscape will continue to change, so we will want to add and adjust techniques as we go.

That’s a deep fake:  Taking a page from martial arts superheroes.   [Defensive]

Sometimes we forget that human beings are astonishingly powerful, creative, and adaptable.  

It is helpful to remind ourselves of real superheroes regularly.   My list differs from day to day.  Today it includes Abraham Lincoln, my niece who literally dances her way through life, and the gymnast Simone Biles.  Three minutes of watching Simone Biles defy gravity lights me up and reminds me what human beings can do.

Some martial arts origin stories suggest that unarmed combat techniques developed as ways to resist oppressive rulers who prohibited ordinary citizens from owning weapons.  Skilled unarmed martial artists can beat an armed opponent and seem almost superhuman.  But human they are.

Humans have many capabilities that few of us use.  Most of us, for example, can feel it when someone is staring at us.  Even without training, we generally know when someone is lying.  We just know.  With training, such as that offered by some military and police forces, we can become very good at detecting lies, even from skilled liars.  We can make use of these faculties.

According to neuroscientists, while our brains receive between six and ten million bits of information a second, we can only consciously pay attention to fifty bits per second.  Most of the information we receive about the world is processed unconsciously.  We can train our conscious and some of our unconscious information processing.  Think muscle memory.  Think intuition. 

Although over time we may want more sophisticated training, we don’t need courses to get started.  As just one example, professors and teachers everywhere are learning on the job to detect AI-written materials without software or AI assistance.   A friend reports that her skills are increasing every week as she pays close attention.

Many of us can detect AI fakes now.  Some are obvious, others less so.  The fakes will get better.  We can get better, too.  We can hone detection skills consciously, using whole-body intelligence and all of our faculties.  It is important that we do so. 

We can use these skills and whole-body intelligence not only to defend ourselves but also to connect and coordinate with other people in positive ways.  Which brings me to a second tactic.  This one calls for positive action.

Are you a Leonardo DaVinci, George Washington Carver, or Marie Curie?  AI can support a world full of polymaths.  [Momentum building]

DaVinci was good at art, anatomy, botany and geology, and engineering.  Carver excelled in agriculture, environmentalism, and art.  Curie had many accomplishments in physics, medicine, and education.

We regard these and other polymaths with admiration and perhaps a little jealousy. 

Many of us can now become polymaths, deeply knowledgeable and creative in more than one area.  We can begin by asking questions that naturally interest us.  And then choose to dig more deeply into two or three areas, increasing our focus and understanding.

We can ask AI to summarize whole libraries of information— each one of which might have taken a lifetime to read.  Like a cookie monster gobbling sweets, AI can gobble digital information of many kinds—books, raw data, images, movies, TV shows, videos, and recorded conversations—and, when instructed, present patterns and potential ways of understanding that data.  Then it is up to us to make meaning out of the data, tell stories, and take action.

Imagine a world full of AI-supported polymaths, each of us able to offer new insights, technologies, and art.  It can be almost like living several lifetimes in one.  And our societies can be greatly enriched.  This widely distributed creativity can be fun and fulfilling and can help balance the power of single-goal institutions.

We also have the opportunity to use AI in ways that may help us restructure our societies to incorporate two things that most of us want:  the opportunity to directly participate in government decisions and the support needed to be happier and to help our families and friends be happier as well.

Better lives:  Using AI to transform ourselves and society.

Click here for Part 3 in this series.

Click here to sign up for our newsletter announcing new posts.

About Us: To supplement rules and laws, the Allison Group’s Social Dynamics framework describes five additional methods of governance, four of which have been broadly tested in history, and one of which is emerging now. Click here for a two-part video introduction to the Social Dynamics framework with an emphasis on regulating AI (scroll down the page).

Governing AI series:  Part 1

This article is Part 1 of a series about the power of using multiple methods of regulating AI.   This is not, for example, about the content of a law or a policy recommendation from an AI developer industry group.  It is about what types of governance have a chance of working and why.

Whether you’re an AI developer, a think-tank specialist, a legislator, or a member of the general public, this is for you.  Understanding how to regulate AI is crucial for everyone.

Fool me once: Shame on you, AKA (also known as) social media

When early social media spaces including Friendster, Myspace, and Facebook began offering digital community space, I was excited about the possibilities for human expression and connection.  I was not alone. 

As is usually the case in new spaces, some experiments prospered and some failed.  Those that prospered offered services we—the general public—wanted.  Use them, we did.  And do.  Accelerating adoption and use rates show an almost insatiable appetite for social media. 

It took us quite a while to see some of the dark sides.  And, even now, we have trouble with it.  Yes, we might think that we should put our phones down more often but seeing pictures, feeling connected, and getting up-to-the-minute news doesn’t feel inherently dangerous. 

But, crushing comparisons of popularity and wealth have led to illness and suicide.  Consumption of unneeded products has destroyed health.  When we are speaking online, digital distance gives us permission to be mean—and worse—to each other. Extreme views and manipulation have destroyed our trust in news and our ability to speak with each other.

Some—maybe all—of these, were unanticipated and unintended consequences. And certainly there are many factors contributing to these situations.

But, as Lord Acton wrote in a letter to Bishop Creighton in 1887, “power corrupts…” especially if not tempered by checks and balances.  Now we know that many social media executives knowingly engaged in practices that increased profits but also caused harm.

Far from checks and balances, in the United States, we have single-pointed laws that require business executives to put short term profits above a more balanced set of goals that include benefits to employees, customers, and others.  Friends across the pond report that, although their business laws do not have single short-term shareholder profit requirements, stockholder expectations and market pressure often result in similar behavior.  (I’ll say more about this in the future.)

Greed for profit and political power drives much of the social media mess we find ourselves in.   Fool me once, shame on you.

Fool me twice:  Shame on me, AKA AI

We are adopting AI even faster and more broadly than we did social media.

Even though there are many warning signs of potential and actual abuse of power by AI developers and users, we are not going to give it up.  I am not arguing that we should.  The benefits are specific and immediate.

However, let’s not be fooled twice.  It behooves us to be skillful in regulating this potent new tool. 

AI and the balance of power

While most AI developers and regulators turn to legislation as the primary means to regulate AI, many have an uneasy feeling that laws won’t be enough.  They are right.

Governments which have individual freedom and protection of citizen and resident rights as an important goal use some form of checks and balances.  Often these come in the form of formal structures and processes, such as the division in responsibility and power between different executive, legislative, and judicial branches of government. 

Checks and balances are only effective when the players are roughly equal in power.  That turned out not to be the case with social media and it is already clear that businesses are the strongest player, by far, in the AI world. 

Trust us.  Fool me twice.

Many AI businesses and executives have good intentions (which they tell us about often).  Some do not and some don’t really care as long as they make money.   Add deliberately aggressive political players to this mix and the need to balance power is very clear. 

Our experience with social media is one example of a time when rules and laws were not sufficient to stop the use of harmful technology. 

Laws and rules are necessary and helpful.  There are good reasons, however, why they are not sufficient.  Here are three of many:

• First, it takes too long to create a useful law; AI technology development won’t wait.
• Second, laws are related to a nation-state and are not binding on those outside its borders, with some wide-ranging international treaties as exceptions.  Even international treaties don’t cover the whole earth, so there are always players on the outside.
• Third, to enforce laws, the policing agency must be able to access and understand the technology well enough to determine whether the law has been followed.  While some AI developers practice open source; others use a closed box for competitive and, perhaps, other reasons.
  
  Even when the code is open, effective regulation requires specialists who understand code that is uniquely and continuously enhanced.  This is not a capacity that is in place now.

To repeat, we will not be able to control AI using laws and regulations alone.

If we don’t wish to be fooled twice, what other options do we have?  I’ll describe several options in this series of articles.  Let’s get started.

One important way to think about AI is to recognize that it is, among other things, a communication technology.  It supercharges communication.  With this in mind, we can look at history to find an example of a time when individual people used a new communication method to dramatically shift the balance of power.

“It doesn’t say that.”  Printed books as a means of balancing power

When the printing press came to Europe, no one had any idea that it would be a major factor in shifting the balance of power from long-established centers of power toward individuals, especially those who learned to read. 

For centuries in most European villages, there was only one book—the Bible—and, at best, only one person who could read, the priest.  Imagine the shock a priest who had been in the habit of stretching the Bible a bit to his own benefit felt when, for the first time, someone stood up in the congregation, pointed to his own Bible, and said “It doesn’t say that.” 

There were, of course, many conscientious and caring priests.  But there were also others less ethical.  The same was true of rulers, secular authorities, and merchants. 

As printing and literacy spread, variations of this scene took place over and over.  

Printing and literacy shook the western world, exposing greed and injustice, and giving more power to educated individuals.  Books and newspapers were a major factor in the Protestant Revolution, triggering a huge redistribution of power away from the Roman Catholic church and toward protestant churches—literally those who were protesting abuses of power and setting up their own religious groups.  

One way of balancing business and government power is to distribute it broadly among citizens and residents, what we might call “the people.” What’s the equivalent of “it doesn’t say that” in the AI ecosystem? 

“That’s a deep fake:” Balancing power in the AI world

Click here for Part 2 in this series.

Click here to sign up for our newsletter announcing new posts.

About Us: To supplement rules and laws, the Allison Group’s Social Dynamics framework describes five additional methods of governance, four of which have been broadly tested in history, and one of which is emerging now. Click here for a two-part video introduction to the Social Dynamics framework with an emphasis on regulating AI (scroll down on the page).