[LINK] Yet another triple-recycled AI Pseudo-Breakthrough

[Roger Clarke]

I've clipped in an article from today, headed:
 > CBA runs AI over 230,000 words of ASIC regulation
 > Australian trial follows similar work in the UK in early 2018.

[That fails to mention multiple working products c. 1990, and c.2000, 
incl. SoftLaw in Canberra.  An email below notes a 2008 recycle.

[I'm reliably informed that Softlaw did *not* parse legislation and turn 
it into code.  The coding depended on inspection and invention by 
skilled people with background in all of law, logic and the composition 
of logical expressions.

[But neither does the 2020 mirageware 'parse legislation and turn it 
into code'.

[The opening para. claims that the new software:
"**helps the bank parse** complex financial regulations and determine 
how best to meet them" (emphasis added).

[And even that's over-claiming.  Elsewhere it says its function is:
"to quickly **identify items in the regulation that could be reviewed 
and actioned** ..." (emphasis added).

[IOW, it's an at best marginal advance on the string-searching tools 
available since the mid-1960s, including to Softlaw 25 years ago.

[AI maintains its long track-record of re-inventing the same thing every 
decade or so, and seldom with much improvement (beyond running on faster 
processors and hence making mistakes more quickly).

[Sorry this isn't expressed in more jolly text;  but Prof. Clerphell 
wasn't readily available when I needed him to exercise his skills.]


CBA runs AI over 230,000 words of ASIC regulation
Australian trial follows similar work in the UK in early 2018.
Ry Crozier
itNews
Mar 23 2020
https://www.itnews.com.au/news/cba-runs-ai-over-230000-words-of-asic-regulation-539687

CBA has run Australian trials of an artificial intelligence tool first 
piloted in the UK to help the bank parse complex financial regulations 
and determine how best to meet them.

The bank revealed the Australian proof-of-concept in a response to a 
senate inquiry into fintech and regtech last week.

"[We conducted] a recent proof of concept ... to understand how we could 
digitise our end-to-end regulatory change detection and management 
processes using an integration between Ascent and the bank’s risk 
management platform,” CBA said.

“The project, completed in partnership with CBA’s risk platform 
provider, IBM, involved over 230,000 words of regulation interpreted and 
converted into a series of bite-size, actionable tasks appropriate for 
the bank, using artificial intelligence and natural language processing.

“[The Australian Securities and Investments Commission] ASIC 
participated as an observer in the project."

iTnews has confirmed it is the same regtech that CBA first trialled in 
the UK in early 2018.

There, CBA used the Markets in Financial Instruments Directive II (MiFID 
II) regulation “as a test case”, running natural language processing 
(NLP) and artificial intelligence (AI) algorithms over 1.5 million 
paragraphs of text “to quickly identify items in the regulation that 
could be reviewed and actioned, saving hundreds of hours of manual 
processing.”

A CBA spokesperson told iTnews that the bank had now run the same 
technology over “two pieces of ASIC legislation”.

“The first was the National Credit Consumer Protection Act, [and] the 
second was ASIC’s Corporations Act,” the spokesperson said.

“These Acts comprised over 230,000-plus words, which Ascent’s artificial 
intelligence and natural language processing capabilities were able to 
ingest, decompose, analyse and convert into bite-size, actionable tasks 
that pertain to processes, policies, systems and reporting for the bank 
within five-and-a-half days from start to finish.”

IBM, which supplies CBA’s governance, risk and compliance platform, were 
a partner to the local proof-of-concept, “co-developing an API with 
Ascent to have data from Ascent’s platform flow into a test instance of 
our governance, risk and compliance platform,” CBA said.

Earlier this month, the bank was named ‘innovator of the year’ by the 
Regtech Association of Australia.

“Our risk and technology teams have been working collaboratively with 
our regtech industry partners to test ways we can use technology to 
deliver better risk, compliance and customer outcomes and they can’t 
wait to take their ideas to the next level,” the bank’s spokesperson said.



-------- Forwarded Message --------
Subject: Re: [LINK] Centrelink relies on artificial intelligence
Date: Mon, 25 Aug 2008 17:21:21 +1000
From: Roger Clarke <Roger.Clarke at xamax.com.au>
To: Link <link at anu.edu.au>

At 17:03 +1000 25/8/08, David Boxall wrote:
 >>  Centrelink is using natural language-based artificial intelligence 
to ensure people get the financial benefits to which they are entitled.
 >>  The intelligence forms the backbone of a business rules and 
compliance software engine created by Aussie developer, Haley.

Haley's predecessor, SoftLaw was doing this right through the 1990s.

For example, a 2-minute Google found a timeline here:
http://bizrules.info/weblog/2007/11/haley_rule_bursts_into_the_bus.html

So what justifies an announcement in 2008, almost two decades after the 
original work?

Note that I'm not expressing doubts about the *substance* of the claims. 
  (I'm an inveterate AI sceptic, and with very good reason. But I've 
never evaluated the SoftLaw work;  and many years back it did seem to be 
making a lot more headway than most AI-based projects).


To follow David's theme then, how about:

'Media fail to use any form of intelligence in vetting media release' ?


-- 
Roger Clarke                            mailto:Roger.Clarke at xamax.com.au
T: +61 2 6288 6916   http://www.xamax.com.au  http://www.rogerclarke.com

Xamax Consultancy Pty Ltd      78 Sidaway St, Chapman ACT 2611 AUSTRALIA 

Visiting Professor in the Faculty of Law            University of N.S.W.
Visiting Professor in Computer Science    Australian National University