MICHAEL STREATFIELD: The jagged frontier where humans and AI work together

From suggesting inappropriate pizza toppings to wiping billions off tech giants’ market caps, generative artificial intelligence’s (AI) early stumbles have been as spectacular as its promises.

It also represents a transformative opportunity in investment management as managers learn to harness the technology’s true potential. While it hasn’t revolutionised the industry yet, tools such as Stanlib Multi-Asset’s bespoke AI tool, MAISY, are refining how we work. The takeaway? Success lies in understanding AI’s limitations while leveraging its strengths.

The journey to harness generative AI reflects the broader industry’s evolution: initial overexcitement followed by inevitable setbacks, settling to practical applications. After early missteps, corporate adoption is on the rise. A Morgan Stanley survey found that 90% of generative AI projects met or exceeded expectations, driven by the promise of enhanced productivity, despite concerns over data security and brand integrity.

Boston Consulting Group deliberately tested the impact of AI access across its workforce and found substantial improvements in task quality for the right-use cases. The big surprise was AI’s ability to act as a “great skill leveller”, helping less experienced workers catch up to their more skilled colleagues. This creates a strong incentive for knowledge workers to adopt AI tools for both upskilling and productivity benefits.

We recognised generative AI’s potential to summarise large amounts of data and support idea generation. However, we also understood the limitations of off-the-shelf language models in generating high-quality investment research. To overcome this, we implemented an approach called Retrieval Augmented Generation, which allows us to boost our AI queries with our curation of proprietary investment data, meeting transcripts and research. The result is our private research chatbot, which aids our investment team’s productivity and information synthesis.

Generative AI provides us with substantial flexibility in output — from tabulating data, contrasting approaches to generating SWOT analyses. It can even play devil’s advocate by presenting counterarguments to challenge assumptions. This ability to consider multiple perspectives leads to more robust investment decisions, which ultimately benefits our clients.

However, we have also learnt that generative AI’s insatiable appetite to “generate” can pose risks. We implemented safeguards to prevent MAISY from producing fictitious information, a phenomenon known as hallucinating. Transparency is another critical issue. We’ve designed the system to reveal the source documents behind MAISY’s responses, enabling our investment professionals to dig deeper into the underlying research.

We’re not alone. Citi has identified three waves of generative AI adoption in investment management: internal models to boost productivity; client-facing tools for communication and recommendations (with human oversight); and eventually “investment co-pilots” emerging to assist portfolio management. This transformation will not happen quickly, as it may require new technology, organisational changes and data consolidation.

There will also be a need for personal adaptation, as investment professionals develop new skills. The Boston Consulting Group found that while AI can enhance certain tasks, it can hinder in areas requiring deep subject knowledge and business experience. Learning to navigate this “jagged frontier” — knowing when to use AI and when to rely on human expertise — will be a critical new competency, as will balancing “how much”, as overreliance on AI can lead to lazy and poor decision-making.

The models themselves are also evolving, with promising developments on the horizon. There’s much anticipation around multi-modal models that can process various input types, both text and images. This could revolutionise how “chart-heavy” investment research is consumed and analysed. Additionally, “reasoning models” are emerging to verify AI-generated answers, aiming to improve reliability and reduce hallucinations. However, these advancements also pose their own challenges, as the “right” answer is often easier to assess in structured domains than in more creative, unstructured work.

From our experiences, we’ve distilled several key lessons for the future:

1. Take the plunge: Start experimenting with AI, even if you’re sceptical. The early mistakes can be entertaining, and the potential rewards are significant.
2. Testing never ends: Remain vigilant, as large language model performance can be inconsistent over time.
3. Trust but verify: Maintain a healthy dose of scepticism and ensure human oversight. Humans need to watch for bias and ethical responses, not just accuracy.
4. Evolve as models evolve: The AI landscape is dynamic, so be prepared to adapt both your technological processes and human skill sets.
5. Human in the loop: Clearly define what should remain under human control. AI can falter, making human evaluation essential.

Futurist Roy Amara was right: “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.” Our experience with MAISY has taught us this truth. While generative AI hasn’t revolutionised investment management overnight — MAISY is certainly not making investment decisions for us — it is a valuable “team player”, steadily refining and transforming how we work.

As we continue this journey, we’re neither AI evangelists nor sceptics, but pragmatic adopters focused on one goal: harnessing this powerful tool to deliver better outcomes for our clients. The future of investment management does not lie in AI replacing human judgment, it lies in finding the sweet spot where technology and human expertise work together.

• Streatfield is senior quantitative strategist at Stanlib Multi-Asset.