Spotting “Probabilistic Parrots”

No, no, he’s not dead, he’s, he’s restin’! Remarkable bird, the Norwegian Blue, idn’it, ay? Beautiful plumage!” – Monty Python

As financial advisers, licensees, and compliance professionals grapple with integrating artificial intelligence (AI) into their operations, or consider whether integration is appropriate, gaining clarity about AI’s true nature and capabilities is a critical governance function. While the term “stochastic parrots,” introduced by Emily Bender and colleagues in their 2021 paper On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?, has provided a useful metaphor, its academic tone can limit practical understanding. That’s why, in recent work, I introduced an alternative conceptual model (“probabilistic parrots“) to describe models like ChatGPT and Claude.

In my view, this reframing provides a clearer and more accessible way to conceptualise how AI language models actually function. Plus, beyond its inherent alliterative appeal, “probabilistic parrots” is more likely to land with most financial services professionals (particularly those non-quants who can’t spell, or even confidently define and pronounce, stochastic). 

Before we proceed too far down this rabbit hole, I’ll acknowledge that “AI” is a broad and worringly imprecise term that encompasses large language models, machine learning classifiers, predictive analytics engines, and natural language processing tools.

These technologies are often unhelpfully bundled under the ‘AI’ label despite functioning and being used in very different ways.

Don’t panic. I’m not offering a technical analysis of the entire industry, but arguing that this subtle rebranding of LLMs like ChatGPT will help us better conceptualise, integrate and manage the impact of this technology. 


My Issue with “Stochastic Parrots”

Assured Support’s on record for emphasising how much language, intent and context matter. It’s no different here. Bender’s original term, “stochastic parrots,” emphasises that despite impressive outputs, large language models like ChatGPT do not genuinely understand meaning. Instead, they generate responses by recognising statistical patterns in their data sets. While the term “stochastic” (meaning random or chance-based) may be effective in tempering expectations of AI capabilities, the term “stochastic” is effectively meaningless for those of us outside academic or technical circles.

For financial advisers and compliance teams, clarity and ease of understanding are critical to effectively assess, adopt and advocate for AI technologies. Effective project management often depends on the effort we make to reduce unnecessary confusion or resistance. This slight tweak might make a huge difference in the “contagious enthusiasm” of some overly optimistic stakeholders. 


Enhancing AI Literacy in Financial Services

I appreciate that you might be concerned that by using “probabilistic”, I’m suggesting that AI systems possess intentional reasoning or genuine intelligence, but the inclusion of the term “parrot” clearly underscores the reality that AI models, like some actuaries, merely mimic human language without genuine comprehension or intentionality. Their responses result solely from complex statistical calculations based on vast amounts of data.

I’m concerned that the financial services industry is in the middle of the AI-hype cycle; where AI integration or AI-solutions are heralded as the answer to each and every inconvenience or impediment facing advisers and licensees. The redefinition recognises AI’s impressive capabilities and its inherent limitations, and provides clarity that will help financial advisers and compliance professionals manage expectations appropriately.

Essentially, adopting the term “probabilistic parrots” reinforces the reality of AI’s probabilistic basis and enables advisers and compliance officers to identify when and why AI-generated outputs may be unreliable or biased. In addition, this understanding should prompt users to more critically evaluate AI-generated content and judiciously use AI technologies effectively without becoming overly reliant on their outputs.


Clearer Communications and Competitive Advantages

As AI technologies become increasingly embedded in financial services, clear and accurate communication about AI capabilities and limitations will differentiate forward-thinking, compliance-focused organisations. Framing LLM as “probabilistic parrots” offers a pragmatic bridge connecting complex AI concepts directly with the practical needs and decision-making frameworks of advisers, licensees, and compliance professionals.

But, beyond that modest goal, adopting the term “probabilistic parrots” is not only a linguistic shift but a crucial step towards greater clarity, informed integration, and confident decision-making.

If you liked this, you might also enjoy reading:

AI in Financial Advice: Efficiency Gains, Compliance Risks, and Cognitive Costs

The Hidden Risks of AI: ASIC’s Review of Licensees’ Embrace of Artificial Intelligence

AI-Driven Compliance: Transforming Risk Management and Regulatory Oversight

The Green Cost of AI: Can AFSLs Balance It All?


Frequently Asked Questions

1. What does the term “probabilistic parrots” mean in the context of AI?

“Probabilistic parrots” is a metaphor that describes how large language models (LLMs) like ChatGPT generate text. These models don’t understand meaning—they predict words based on statistical patterns in massive datasets. This framing helps non-technical professionals, especially in financial services, grasp that AI tools mimic human language without genuine comprehension.

2. Why is the term “stochastic parrots” considered less effective for financial services professionals?

While “stochastic parrots” accurately reflects the statistical nature of LLMs, the term “stochastic” is unfamiliar and confusing to many outside academic or data science circles. Financial advisers and compliance teams benefit from clearer, more intuitive language like “probabilistic parrots” to better assess and integrate AI tools into their workflows.

3. How can financial advisers use the “probabilistic parrots” model to improve AI governance?

Understanding AI as “probabilistic parrots” promotes critical thinking about AI outputs, emphasising the need for human oversight. This model supports better risk management, encourages scepticism of AI-generated content, and helps compliance teams set realistic expectations when adopting AI-driven solutions.

4. Does using AI in financial services pose risks due to its probabilistic nature?

Yes. Since AI models rely on statistical predictions rather than comprehension, their outputs can be unreliable or biased. Recognising their probabilistic nature helps financial professionals identify potential inaccuracies, reduce over-reliance, and apply AI tools more judiciously in decision-making and compliance processes.

5. How does adopting the term “probabilistic parrots” give firms a competitive advantage?

By communicating AI’s capabilities and limitations more clearly, firms can foster smarter adoption, improve internal understanding, and manage client expectations. This more accessible term was coined by Sean Graham to strengthen trust, aid compliance, and position firms as forward-thinking leaders in AI governance and integration within the financial sector.

Keep exploring

Spotting “Probabilistic Parrots”

Subscribe

Every fortnight “Three Hit Tuesday” delivers thought leadership, considered analysis and insights that will help you improve your advice, more effectively manage your regulatory risks and make you better informed than your peers.

AS-Subscribe Form

"*" indicates required fields

This field is for validation purposes and should be left unchanged.

We respect your privacy. We know everyone says that, but we promise that we won’t sell your contact details to dodgy telemarketers, spam your email or otherwise exploit your trust.

Step 1 of 8 - Your Role

This field is for validation purposes and should be left unchanged.

Assess your ASIC exposure

Answer a few targeted questions to identify where your compliance may not stand up under ASIC review.

Takes less than 2 minutes. No preparation required.

What best describes your role?