> ## Content Index
> Fetch the complete content index at: https://www.drdanielcash.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# From Mauritius #3: What If the Machine Is Better?
- URL: https://www.drdanielcash.com/from-mauritius-3-what-if-the-machine-is-better/
- Published: 2026-10-06T16:59:17.000Z
- Updated: 2026-10-06T16:59:17.000Z
- Author: Daniel Cash
- Tags: Insights, Credit Ratings

The question came near the end of the panel that I was on in the second day of the 2nd Annual Conference on Credit Ratings in Mauritius. I am paraphrasing, but it went roughly like this: given how quickly AI models are improving and how great artificial intelligence may become, will credit rating agencies simply be redundant in ten years?

I said no. I said it without much hesitation, and I gave reasons I still believe. But I had walked into that session expecting a different conversation. The panel was on AI and the credit rating industry, and I had assumed we would spend the time on incorporation: where machine learning might sit in the analytical process, what it could do for surveillance, how a methodology changes when part of the work is done by a model and, ultimately, what the governance implications are (where I am most interested). The question from the floor skipped past all of that and I was glad, because it really made me think. It asked whether the institution doing the incorporating might eventually become unnecessary. I answered it, and then spent the evening (and a fair part of this morning) wondering whether I had answered it too quickly.

What follows is a thought experiment. I am not predicting anything. I want to take both answers seriously, beginning with the one I did not give.

Imagine AI systems becoming extraordinarily capable at credit analysis. Arguably, they already can, but stick with me. They ingest financial statements, fiscal data, central bank communications, election results, commodity prices and bond spreads, and they process all of it continuously rather than at the rhythm of a committee calendar. They compare issuers across jurisdictions without fatigue or the quiet preferences that every analyst develops over a career. They find relationships between variables that nobody would have thought to test, run thousands of scenarios, and produce an assessment more quickly, more cheaply and perhaps, eventually, more accurately than a team of analysts.

If that happens, what precisely remains valuable about having a credit rating agency produce the assessment? I find this harder to dismiss than my answer on the panel might have suggested. The analytical work at the centre of the rating process is labour-intensive, expensive and slow, and those are exactly the characteristics automation tends to erode. It is conceivable that the analysis becomes progressively automated and then, as these things usually do, commoditised. My argument focused on my 8th book which presented the rating systems of credit and ESG rating agencies as signal providers, and my forthcoming 10th book which focuses precisely on the credit rating committee and concludes that it is, for varying reasons and one of which is its human element, a key source of the agencies’ authority to even rate at all.

This led me to think about the direction the major groups have already travelled. Over the past two decades they have built well beyond the traditional ratings business, into data, analytics, research and the infrastructure that sits underneath financial decisions. I do not think they did this because they expect ratings to disappear, and I have seen nothing to suggest it. But I do wonder whether that diversification leaves them particularly well placed if the economic value shifts away from producing a rating and towards supplying the data and analytical machinery on which evaluations are built. In that future the groups could remain enormously important, perhaps more so than now, while the credit rating itself becomes a smaller part of why they matter.

Then there is the case for no, which was my answer.

My reasoning was that a credit rating agency does something more than calculate credit risk. A Big Three representative in the audience described the agency as an intermediary, and I found that a useful word. Countries and companies speak to the credit rating agency. Analysts question them, sometimes uncomfortably. The agency then speaks to investors. There is interaction around the judgement, and the agency sits inside a set of relationships between issuers, investors, and markets. A judgement formed through that kind of exchange is a different object from a number produced from data, even when the two arrive at the same place.

My own way into this is ‘evaluative authority’. A credit rating is not valuable simply because somebody has calculated something correctly. It is a signal issued by an institution whose judgement has acquired authority over time. That authority is partly historical, partly reputational, partly embedded in market practice (in contracts, investment mandates, the habits of portfolio managers), and still human-centred. A committee voted. People can be asked to explain themselves either within the processes, or as an organisation to regulators later down the line (it is worth me noting, importantly, that individuals are not brought to task for their opinions outside of the organisation, and for several important reasons).

So consider a simple case. An advanced AI system assesses a sovereign as BBB-. S&P Global’s rating committee independently assesses the same sovereign as BBB-. The output is identical. Are the two judgements equivalent?

My instinct says they are not, although I struggle to say why without reaching for language that sounds like sentiment. So I pushed it further. Suppose the AI system is better. Suppose that over a decade it proves more accurate at estimating credit risk than the human-centred process. Would markets transfer authority to it?

I am honestly unsure. Accuracy matters, obviously, but authority does not seem to follow accuracy in any automatic way, and the gap between being good at evaluation and possessing the authority to evaluate is the thing I have not been able to stop thinking about since the session ended.

Part of the answer, I suspect, lies in coordination. Credit rating agencies supply information, but they also supply a recognised signal around which other actors organise. Investors know the judgement. Issuers know it. Regulators may recognise it, and every participant knows that everyone else has received the same signal. A more accurate assessment held privately by one fund is useful to that fund, but it cannot do that work. This makes the AI question considerably more complicated than whether machines will eventually outperform analysts.

I can see several ways it might go. Human judgement may simply remain essential. Agencies may automate more and more of the analysis while preserving a human and institutional judgement at the end, which is roughly where the industry seems to be heading now. Traditional ratings may lose economic weight while the parent groups grow in importance as data and analytical infrastructure providers. Or machine-generated evaluation may itself begin accumulating the authority we currently associate with established institutions.

That last possibility is the one I keep returning to. If investors, issuers, and markets eventually recognise and organise themselves around machine-generated assessments, the machine will have moved beyond being good at analysis and begun to acquire evaluative authority of its own. I do not know what that would look like in practice. Who would the finance ministry speak to before the assessment was published? Who would answer for it when it was wrong? Perhaps markets would decide they did not need anyone to answer.

I told the audience member no, and I still think that answer rests on strong foundations. But the question stayed with me because it showed that AI in credit ratings may eventually raise something deeper than which parts of the rating process can be automated. It may force us to ask what a credit rating agency is actually for.