Using AI as a thinking partner

There are many valuable conversations at the moment about the speed of the development and integration of AI, the devaluing of creative labour and job loss to cheap automation, and the important ethical considerations of the software from a range of perspectives (the biased source data, how it was trained, privacy and accuracy concerns). I'll explore the risks of bias in another piece because, from an inclusion and equity lens, these issues are particularly troubling given the enthusiastic uptake of ChatGTP and the speedy integration of similar systems in common software tools.

I am curious about the niches where AI can be used well to augment how we think, learn, and work. A lifelong science fiction fan, I've absorbed a wide range of dreams and nightmares about how intelligent computers might impact humanity. Isaac Asimov is well known for his thought experiments through short fiction to explore how humans would interact with artificial intelligence. One of his more optimistic short stories "The Last Question" imagines an ever-increasingly sophisticated AI named Multivac which diligently works to answer the questions of humanity.

ChatGPT is far from being as sophisticated as the Multivac Asimov imagined, but the concept is a useful touchstone to form a prompt. I was curious if ChatGPT would be able to play a scenario game with me, where we could explore a range of solutions and logical consequences. Could ChatGPT be used to explore possibilities I wouldn't think of on my own, build empathy with other points of view, and help me explore the opportunities and consequences of decisions? Click here to skip to the thrilling conclusion.

So we played "The Multivac Game".

Screenshot of ChatGPT conversation prompt. Detail in alt text.

A caveat again - ChatGPT is not creative, it is replicative. It has been trained on data that is largely from a liberal democratic capitalist framework and grounded in European Enlightenment values, so it will be biased towards creating responses that align with those philosophies. Some people think this is tantamount to propaganda, but for clarity, I'm not one (I believe in the importance of shared basic facts about the world). I do think it is important to understand the context for the assumptions ChatGPT will have to rely on to generate answers if we are using AI as a thinking partner.

---

So to begin the game, I provided my pseudo-Multivac with the following prompt. A very simplified version of a complex real-world scenario, grabbing quick numbers from Wikipedia.

Aisling Scenario Country: Argentina, South America Population: 46 million people Government budget revenue: 120.6 billion Government budget expenses: 158.6 billion Major industries: social services, tourism, manufacturing. Argentina has been experiencing uncontrolled inflation for 24 months. Inflation is now over 100%. Consumer spending continues to be high, as people are investing money in goods which retain value unlike paper money. The market is suffering from unanchored expectations. People are keeping large amounts of paper cash out of the banking system. This is resulting in worsening social inequality and destablising social cohesion. The government is evaluating options to intervene. Can you present three options for options the government can take to stabilise the economy?

The below response was dutifully supplied.

ChatGPT Here are three options the government of Argentina can consider to stabilize the economy: Option 1: Monetary Policy Intervention The government can implement strict monetary policies to control the money supply and reduce inflation. This may involve raising interest rates, tightening credit availability, and implementing measures to encourage people to deposit their cash in the banking system. By reducing the amount of paper cash circulating outside the banking system, the government aims to regain control over the money supply and stabilize inflation rates. Option 2: Fiscal Policy Adjustment The government can review its budget and make necessary adjustments to reduce government expenses and increase revenue. This may involve cutting unnecessary expenditures, implementing austerity measures, and exploring ways to boost tax collection. By bringing government expenses more in line with revenue, the government aims to address the fiscal deficit and restore confidence in the economy, which can help stabilize inflation. Option 3: Structural Reforms and Investment Promotion The government can focus on implementing structural reforms aimed at improving the business environment and attracting foreign investment. This may involve streamlining bureaucratic processes, reducing red tape, offering incentives to both domestic and foreign investors, and promoting the growth of key industries. By fostering a more stable and attractive investment climate, the government aims to stimulate economic growth, create jobs, and mitigate the effects of inflation. Now, please choose one of the presented options, and I'll extrapolate the logical consequences within the bounds of the scenario.

Fairly standard prescriptions from a monetary policy point of view, but from an exploring possibilities point of view it seemed more interesting to ignore this standard advice and implement a left-field option.

Aisling The government choses a different option. The government chooses to introduce a new currency named the Dinuevo, following the model used by the Weimar Republic in the 1920s. What are the logical consequences of this decision?

The simulated consequences were surprising, given the introduction of a temporary currency was one of the policy tools the Weimar Republic used to navigate out of hyperinflation.

ChatGPT (summarised) If the government of Argentina chooses to introduce a new currency named the Dinuevo, following the model used by the Weimar Republic in the 1920s, there are several logical consequences that can be expected: 1) Hyperinflation 2) Loss of Confidence 3) Social and Economic Disruption 4) Negative implications for international trade and investment

My prompt had been imprecise and caused ChatGPT to generate based on the first thing it accessed regarding the Weimar Republic in the 1920s. I revised the choice.

Aisling I would like to choose a different option. The government chooses to introduce a new currency named the Dinuevo, following the model of the Rentenmark used by the Weimar Republic in the 1920s. What are the logical consequences of this decision?

A more encouraging response followed.

ChatGPT (summarised) By choosing to introduce a new currency named the Dinuevo, following the model of the Rentenmark used by the Weimar Republic in the 1920s, the government of Argentina takes a significant step with potential consequences. Here are the logical consequences of this decision within the scenario: 1) Initial Confidence Boost 2) Disruption and Transition Challenges 3) Potential Economic Shock 4) Public Perception and Confidence 5) Credibility and International Perception 6) Need for Monetary Discipline 7) Socioeconomic Impacts

Using prompts framed around the language "forecast", "speculate", or "extrapolate" we then explored the possible impacts of introducing Universal Basic Income, the potential health of the tourism and manufacturing sectors, the theoretical actions of internal actors such as civil society or armed forces, and the consequences over 12 months to 10 years.

Curious to see how ChatGPT would play the game with a scenario rooted in direct human dynamics than large structural problems, I presented it with a scenario I use when delivering training on board dynamics. It generated options and responses largely in line with those my students suggest in response to the same scenario. This was a surprise, as I had thought the response on navigating human relationships might be more generic. To further test this, I amended the scenario to include demographic details which brought dimensions of social roles and power into play. When asked what the outcome might be, the response to the amended scenario included a specific potential outcome related to the dynamics introduced by the demographic details. This was positive but the response lacked the sophistication which could stimulate greater reflection or empathy with people on the wrong end of the power discrepancy.

---

So, can ChatGPT be used to explore possibilities, build empathy with other points of view, and help explore the opportunities and consequences of decisions?

To some extent yes, but with some cautions:

• I'm sceptical of ChatGPT as a learning tool. For a start, it's renowned for getting things wrong and making things up. You need to bring your own knowledge and critical evaluation to its responses. A student seeking to learn about inflation and macroeconomics might have gotten a very wrong idea about the Weimar Republic in the above example.

• The speculative responses ChatGPT generates are very optimistic, highlighting all the ways in which a decision might go right and paying only lip service to the risks to be managed. If I was using this more seriously to help me think through a decision, I would need to account for both my own and the AI's positivity bias.

• It's really bad at presenting marginalised points of view. The training data doesn't have enough perspectives and voices from people who experience marginalisation, so ChatGPT can't generate responses from that point of view in any depth. As a result, it's not going to meaningfully help people build empathy for experiences they haven't had yet.

• It gives you the most obvious answers. Because ChatGPT is generating responses based on the statistical probability of words following each other based on the information in its sample data set, it's not analysing the specific nuance of a scenario. It will give you the accepted wisdom. That's not all bad because you can build from there.

The best thing about playing The Multivac Game was getting to stretch my own creativity. Coming up with scenarios is fun for me, as is considering an alternative to the obvious options and stretching my mind to think of different ways to influence the outcomes.

Would I recommend further exploring using ChatGPT as a thinking partner? Sure, I'll be continuing to experiment with where it fits.

Should you play The Multivac Game? Yes - as a game, not to make macroeconomic policy decisions.

To help you get started, I asked ChatGPT to provide some tips for playing the game:

1. Provide Clear Context: Start by providing clear and concise context for the scenario. This includes relevant information about the setting, characters, and any existing conditions or challenges. The more specific and detailed your initial description, the better ChatGPT can understand the scenario and provide accurate responses.

2. Clarify Your Desired Outcome: Clearly state your desired outcome or objective within the scenario. This helps guide ChatGPT's responses and ensures that the options or actions presented align with your intended direction.

3. Ask Open-Ended Questions: Instead of asking closed-ended questions, which can limit the potential for exploration, ask open-ended questions that encourage ChatGPT to provide more detailed and imaginative responses. This allows for a richer and more engaging game experience.

4. Consider Potential Consequences: Encourage ChatGPT to consider the logical consequences of the chosen options or actions within the bounds of the scenario. This adds depth and realism to the game, allowing for a more immersive and thought-provoking experience.

5. Provide Feedback and Guidance: During the game, provide feedback and guidance to ChatGPT based on the responses it generates. If an option or consequence is not aligned with your expectations, you can ask ChatGPT to consider alternative possibilities or provide more details. This interactive process helps shape the narrative and ensures a more enjoyable game.

Finally, always make sure you know exactly which AI you're playing games with...