When AI stops hallucinating, where will that leave consultants?

IT HAS been a gratifying time for people who delight in denouncing “artificial intelligence (AI) slop”, especially since Deloitte Australia recently handed them yet more ammunition. The consulting firm has to partially refund the Australian government for work done on a A$440,000 (S$370,000) report, after an academic found that the work contained multiple errors caused by the use of AI. 

According to The Australian Financial Review, the report, commissioned for the Department of Employment and Workplace Relations (DEWR), included non-existent academic references and a “made-up quote” from a court judgment. 

The report has since been corrected, and the DEWR has maintained that the errors had no bearing on the report’s substance and recommendations. 

Even so, Australia’s lawmakers were unimpressed. “Deloitte has a human intelligence problem. This would be laughable if it wasn’t so lamentable,” Labour Senator Deborah O’Neill said last week. 

“Anyone looking to contract these firms should be asking exactly who is doing the work they are paying for, and having that expertise and no AI use verified. Otherwise, perhaps instead of a big consulting firm, procurers would be better off signing up for a ChatGPT subscription.”

The senator makes an intriguing point, however inadvertently. Why hire an external party under a shared delusion of purely human work, when you could use a large language model like ChatGPT and therefore remove all doubt? 

For both Deloitte and Australia, this gaffe was more the fault of man than machine – a failure on Deloitte’s part to originally disclose usage of generative AI and verify the output, and a failure on the government department’s part to give the finished product a once-over before releasing it to the public. 

This incident represents an uncomfortable dilemma for consulting companies. Do these firms proactively disclose their AI usage and position themselves as experts in AI-augmented analysis? That sufficiently covers everybody’s rear ends, but might impact their ability to command a fee premium, since some of the work is being done by AI. 

Or do consultants take the artisanal route, avoiding AI involvement entirely and market their unadulterated human judgment, honed by decades of experience? That approach, however, presents problems of efficiency and scale, as their competitors use AI to churn out comparable work in less time or for less money. 

At some point, the consulting industry might not even have the luxury of those two choices as the error and idea gaps continue to narrow. AI systems that once struggled with basic analysis are already handling more sophisticated analytical work – the kind on which consulting firms have long staked their reputations. 

And just as the laws of physics are being violated less and less in AI-generated videos every month, kinks in learning and reference will continue to be ironed out at an inexorable pace. 

For now, it is clear what Deloitte Australia should have done when using a tool prone to hallucination. It still had a role to play in wielding AI less ham-fistedly and being a better guardian of its finished work. 

But in the future, it is much less clear what Deloitte and its peers will do when AI models no longer hallucinate, and can generate ideas of comparable rigour and originality, enabling clients to produce a similar outcome on their own. Now, that is the A$440,000 question.