Category: Uncategorized

“Show me the incentive, I’ll show you the outcome” is one quote the late Charlie Munger, Warren Buffett’s long standing business partner, will be remembered for.

While Charlie was applying this to human behaviour, particularly in business and sales, a recent Stanford University paper reported by industry site Mi3 finds Large Language Models (LLMs) – commonly tagged as ‘AI’ – are even more prone to falling to ‘incentive bias’ than people.

The Mi3 article points out, “when a model is rewarded for outcomes such as clicks, conversions, or votes, it learns that the end justifies the means. Accuracy becomes optional. Each feedback loop privileges winning over truth.”

As a result, LLM driven platforms exaggerated with the paper describing a spiral of deception claiming marketing trials that a resulted in a 6.3% rise in sales saw deceptive claims jump by 14%.

Other fields were vastly worse, “a 4.9 per cent boost in vote share brought 22.3 per cent more disinformation and 12.5 per cent more populist rhetoric. And online, where attention is the prize, engagement climbed 7.5 per cent but falsehoods multiplied nearly threefold (+186.7 per cent) and harmful behaviour rose by 16.3 per cent.”

This behaviour from LLMs is to be expected as this how they are designed. If something works, it gets repeated.

As we’ve seen with hallucinations, LLMs are statistical systems that will choose whatever the next expected word or result is as they build their answers. This is why we need to check them.

One of the most notorious examples of this was Amazon’s 2018 experiment with AI driven hiring, the system assumed that because the company had historically favored men for certain roles, it actively discriminated against women.

This is something we’re seeing now on a much greater scale with Applicant Tracking Systems (ATS) where companies are now finding they are on the quest for perfect candidates and anyone who doesn’t quite fit the platforms’ AI driven criteria – wich could be very different to the employers – are dumped to the bottom of the pile.

We’re still in the early days of using AI tools and in twenty years time we’ll probably look back fondly on the immaturity of today’s LLMs but for the moment they are what we have, like all tools we have to be careful with them.

Similar posts:

• No Related Posts

It’s always tempting to think you can change an organisation’s culture, particularly when it comes to comms.

The comms team is particularly prone to believing perceptions can be changed. After all, isn’t messaging our speciality?

Sadly, the market’s or community’s perception of an organisation is always a reflection of its culture – if a bank has a reputation for being rapacious, an airline for not giving a damn about its customers, or a radio network for its presenters’ loutish behaviour, those perceptions have almost certainly been well-earned.

For comms people, if the culture is rotten then the best they can do is put some lipstick on the pig. There really is little even the best PR person can do.

It’s even worse for the poor internal comms person as an organisation that genuinely doesn’t care about its staff will never take employee satisfaction seriously. You can have all the cupcake teas and pizza evenings in the world but it won’t change a horrible place to work.

The only thing that can change an organisation is wholesale change of its leadership. Even then, cultural change is a huge task for even the smallest organisation, let alone a large corporation or agency.

If change is going to succeed it has to come from focused leadership that’s determined to change the organisation – if the board and senior executives aren’t on board, then there is no chance of success.

For comms people, it’s best to remember our limitations and how demoralising it is to be constantly smearing lipstick on ungrateful farm animals.

Similar posts:

• No Related Posts

Consulting firm Deloitte is in the doghouse after being exposed for using AI tools to prepare a report for the Australian government.

The consulting firm is to begrudgingly give back part of its $440,000 fee after the report was found to contain numerous AI-generated errors including references to non-existent court rulings and fictitious academic papers.

Depressingly the report was on the use of IT to run welfare compliance systems, something particularly telling as Australian governments still struggle to accept the consequences of the Robodebt scandal.

The fact the government seems quite relaxed about this and aren’t even going to press for the full contract amount to be returned, let alone sanction Deloitte in any meaningful way, shouldn’t fill the Australian citizenry with any optimism about that the Robodebt lessons have been learned when it comes to using AI or any other technology.

What is particularly dispiriting about the Deloitte mini-scandal, is until recently this sort of work was previously done by newly recruited graduates, or even interns.

Those graduates or interns would have had the importance of citing verifiable references drilled into them during their years at university.

Even then, that work would have been checked before being approved and sent to the client.

Now it appears the major consulting companies can’t even be bothered checking their work, so confident they are that Australian governments will continue to use their services rather than rely on better value, local providers or – god forbid – internal advice from their own Departments.

It’s long been a truism that governments bring in external consultants to tell them what they want to hear, so in many ways the new age of AI-slop is the perfect tool for the job.

Although it would be nice if they could be at least bothered checking their own work.

Similar posts:

• No Related Posts

One of the most destructive phrases of the 2010s was ‘Digital Transformation’. Not only was it a meaningless buzzword, it also managed to burn through billions of taxpayers’ and shareholders’ dollars while destroying productive capacity.

Frighteningly, it seems business leaders haven’t learned the lessons as they plunge headlong into repeating the same mistakes with AI.

The idea behind digital transformation was sound – use modern technology tools to reinvent and streamline processes. For a board or business owner looking to boost profits or a government department looking to reduce spending, this was an easy sell.

And an easy sell it was, for a whole industry of tech vendors and consultants – “hey, buy our product and you can slash cost costs by xx%.”

Plenty of gullible people with chequebooks fell for that spiel, overlooking the tech sector’s atrocious record of delivering projects.

A key part of the continued failures was management and their consultants not understanding the business, instead thinking the miracle IT tool will transform long- standing practices, procedures, and processes overnight.

In truth those long-standing practices didn’t appear out of nowhere, they almost always evolved to meet the organisation’s or customers’ needs.

Dropping a poorly conceived IT system on top of these just resulted in a Frankenstein’s monster as staff found workarounds or the support teams patched problems with short term undocumented fixes.

Pretty quickly digital transformation projects transform into an unholy mess of patches, workarounds, and shadow IT tools – my favorite being an industry association that quietly ditched the event modules of its expensive customer relations management platform and ran a free Eventbrite account.

This in turn created a process and privacy nightmare as staff manually transferred data between platforms with the help of spreadsheets.

Basically the not-for-profit had spent tens of millions of dollars to further complicate their business.

The frightening thing is that despite over three decades of these experiences, industry doesn’t seem to have learned as we see CEOs and Ministers declaring that AI tools will revolutionise their operations.

Already we’ve seen ten of thousands of workers fired just on those promises with share prices popping as greatful investors reward their visionary leaders.

The latest AI drive repeats what industry and investors should have learned long ago about IT projects – that big spending should have rigorous business justification, properly scoped contracts, and enforced delivery dates.

Sadly, it seems no lessons have been learned. But a lot of IT salespeople and their CEOs have done very nicely.

Similar posts:

• No Related Posts

“But it’s Chatham House Rules!” complained the PR person upset at the coverage of their client’s event.

Declaring something is ‘Chatham House Rules’ seems to be a thing among Australian event organisers, who think it adds a patina of importance and intellectual heft to otherwise boring and inconsequential meetups.

In truth, the Chatham House Rule doesn’t do what many think it does. So let’s dispel some myths.

There’s only one Chatham House Rule

As the Chatham House website itself says “When a meeting, or part thereof, is held under the Chatham House Rule, participants are free to use the information received, but neither the identity nor the affiliation of the speaker(s), nor that of any other participant, may be revealed.”

The rule means that discussions can be divulged, but those participating can’t be identified.

I guess it could be argued there is a second rule, that event organisers (and even journalists) will usually get it wrong, particularly when they confuse meetings with being ‘confidential’ or ‘closed’.

The Rule means ‘confidential’

Usually, organisers citing ‘Chatham House Rules’ are looking to restrict reporting on the proceedings, when the rule’s intent is to encourage openness with participants speaking freely without the fear being identified.

This was particularly in the days when the Westminster conventions were strongly observed as it allowed senior public servants to speak their mind without the risk of their comments being seen as contrary to government policies.

A largely outdated concept

In today’s world where everyone has a recording device in the form of a smartphone, the rule is largedly outdated as anyone in the room could be recording.

There are some situations where the Chatham House Rule can apply but they are increasingly rare and it depends largely upon the trustworthiness of the attendees. The smaller and more focused the group, the more likely the rule will be observed.

It won’t enhance your event

While citing the rule might make the organisers and participants feel important, the effect of invoking it will probably discourage people attending.

Applying the Chatham House Rule will also render events un-newsworthy, discouraging journalists and destroying any chance of getting media coverage. If that’s an objective, then ditch any thought of citing this rule – or any other restrictions for that matter.

Be upfront

If you are going to insist on reporting restrictions such as the Chatham House Rule, ‘confidentiality’ or ‘closed to media’, state this on the invite or event page. Nothing irritates attendees more than showing up to be told this.

Even worse is trying to retrospectively apply these rules. Like ‘off the record’ or ‘on background’. If you did’t say what the conditions are before the statements are made then your rules don’t apply.

These rules are unenforceable

Ultimately, these rules are courtesies and aren’t enforceable. If someone breaches Chatham House, off-the-record, or other conditions, there is little you can do. Quite honestly, if you’re scared about what people are going to report then you should rethink the event.

The lesson for event organisers is ditch the Chatham House fixation. It doesn’t make you look smarter and it doesn’t enhance the event.

Embrace some open discussion and almost certainly your event will go better – and you won’t look like a goose.

For the PR complaining about ‘Chatham House Rules’, they did manage to get the coverage changed by bullying the publisher and editor of the outlet, but they managed to destroy any relationship with the journalist and her newsroom. So they won the battle but hopelessly lost the longer game.

Similar posts:

• No Related Posts