Author: Maverisk

Apologies for the play on the pronunciation of Auditing AI for Camouflage. Auditing AI for Stealth would’ve been taken to abbreviated with two s’es, winkwink right?

And I don’t mean this sort of training. Or maybe I do:
How are adversarial examples in AI, not what previously was called camouflage ..? Hence:
Why aren’t camouflage techniques, or rather anti-~, used in training AI systems, and in auditing them for quality of operational survivability ..?

E.g., through the realisation that camouflage is really, really effective when one would want to pick off insurance claims from hapless auto-drivers, a tautology (like here, similis).

And, consider that this area of camou-vs-anti, is old hat.
Weren’t the ‘original’ F-117 stealth bombers not so much, through flipping ..? [As here; just can’t find a link to the rumour that already at operational release, Russian 50s radars kept in reserve behind the Ural, could pick off F117s with much ease.] And Mk1 eyeballs already helped a bit, too, and still. When the technology concerned was developed, wasn’t it overly focused on the particular strand of arms’ race, unaware of the huge context hence going to the (math-wise) limit of the race to nowhere while forgetting other ‘basics’?
And what were IEDs other than stealthed explosives?
Apparently, the lessons haven’t been learned. Lessons will be repeated until they are learned. In this case, at the cost of how many lives, and sorrow on relatives? And how much cost to society (through not spending the ginormous investments on bettering the fate of the underprivileged)..?
Plus, the AI field is (going to be) deployed as a new battle area, apparently.

Now, the focus is on protecting pedestrians against auto-cars, even when protective security is still a problem, possibly not categorically solvable.
Where training is focused on picking out pedestrians, to avoid (or ..? Apparently, there was a no-breaking systems overrule since avoiding fender damage from too close up back vehicles was more important). And adversarial examples are all around. Still, to poke fun (really …!?) and to learn.

But class solutions may not be. Case in point: It’s all still about the above camouflage/stealth issues, the same as ever before. When humans were enlisted to see through it, things not necessarily worked out (just an example). Though simple stuff, did work.

So, you better learn how crypsis and mimiry work.
Because, question: Why did the pedestrian cross the road? To catch insurance claims? Where else do you deploy AI systems, and may encounter such ‘pedestrians’ in analysed traffic/environments?

Maybe complex hybrid AI systems are needed, required. Like, some that do pay attention to secondary (Kohonen) classifications.
Whereas the above camouflage pics showed, humans had only been able to recognise the obvious, to survive the savannah (hey predator camouflage was also just good enough to work a little bit not more given evolutionary development costs), systems learning from scratch may need to be bigger than our brains. Against insurance claims, as that seems to be a more economical (evo, too; evo favours the economical over the ethical …), but also in other fields, against deceit on purpose or not.

Now, I haven’t given you the answer on how to audit your AI system for such (high) quality. Duh – I’m only approaching a publication (date) and it could be my daily bread. Hire me and I can tell ;-/

Oh, plus:

[Already here, it works a bit (when seen through your brows); Twente AFB again, analog pic again]

Oh and to close out, a recent find:

Yet another reminder that Real risk management may not be too different from management.
In that, as said before (here), one would need to start at the start to make progress, like changing the ones responsible, through making them responsible.

Which cannot be done other than through job descriptions. Like this one, the reminder.

So, there is something brewing here and there. Mostly there, and not mostly at that.
Why oh why can’t we seem to get this done ..? So that e.g., the 3LoD thing can be put to rest ..? That was described here already.

I’ll keep on hoping. Hope is what’s left when arguments have run out …?

And:

[Once were city castles the management of risks; Nancy]

Well, well, we we at it again. An On The Job assessment of recruiting-HR. If you wondered: It didn’t go well – for the also present possible/prospective chief [hereafter: chief] with a somewhat-urgent vacancy. The following may read as accusative to the feeble-hearted reader, but is not meant to be other than an ironic display; could fit in a Monty Python sketch…

What happen? Well, it was a throw-back to times before that became a meme. If anything of the previous two sentences doesn’t make sense, then well, there you have it.
Last time I had a similar experience, at least was five years back… (as here), not now that we’re closing in on the 20-20s. Dinos seem to have a knack of staying around.

Which also is a summary of – in my experience of 45mins.; maybe your mileage varies – the capabilities of the recruiter. [With which I mean the HR guy [hereafter: guy], not the external liaison agency.]
This actually being the guy that interviewed me six years before, with similar quality, with similar results. Not that at that time, the recruiter had any issues to communicate, only an apparent grudge for reasons [truly] unknown to me. This one then, hadn’t developed over those six years, either professionally or personally:

At the reception, I met the chief. We had both worked for some other firm (close but not together) more than a decade earlier, but still were on the same communications level immediately. When she was off for coffee, the guy asked whether I’ve been at the office before. I said I had been there times immemorial ago [see above, I had checked my emails from back then in advance about how it went, then], the guy didn’t seem to recall.

Then, of course there were questions, repeated, about why I might consider a perm contract while having been independent for over seven years. I explained, truthfully. Didn’t want to also bring up this directly and/or in flip side, as the guy might feel implicated. No (as in zero) questions beyond what any uninitiated could read from my CV qua run-off-the-mill work content, from the guy. [Chief did ask, and we had a good, somewhat productive even, back-and-forth on content and developments therein. See positive.2.a-.b below. Though I had become less sure that the co. would provide a good environment, open to current-day and future methodologies [e.g., zero from their side on algo trading] and internal organisation market standards. Relying on 3LoD isn’t what one hopes to find w/ a new employer… Also, the privacy officer part of the position (yes) was also discussed. Apparently, instead of independence v.v. operational privacy tasks, subsumption under-under-… in the ‘2nd Line’ was intended. Not quite up-to-date with jurisprudence, eh?] A couple of vague ‘STAR’ attempts followed; how’zat for outdated/never-worked interview things…

In return, I asked about the position, of course. To which the chief answered; not the guy too much. At some point, we came to working hours. Though the industry concerned has over two decades of experience with 36-hour work weeks as standard, the guy was vehemently defending (?) 40-hour contracts, with a need to be flexible around that, in the plus definitely not the minus. Not within that even, as is customary everywhere in the country. I have no indication otherwise than that the assumption with the guy was that employees working 40 hours, actually are productive for 40 hours not 10 at most as is scientifically established, or that flex keeps one at minimum levels of effectiveness through being able to pick up the tiniest bits of development qua trade and content let alone development for one’s (professional) career. Or that being able to represent the co. externally, in professional-trade circles, might enhance future recruitment efforts.
Oh and of course, required salary was asked. I gave a ballpark, based on what I know to be a perfectly normal 12-months figure ex.benefits for this level/scope of position, he apparently took it as a final full-package ask price. Uhm, how did this and this [both Dutch] come about ..?

Next to that, my secondary engagements for which I’d wish to keep some flex and hours, and which the chief recognised as important for sane and proper future functioning as a professional, would be impossible to keep up according to the guy. To the point that a three-day teaching engagement two and a half months ahead, would make it impossible to on-board before that; zero chance. The guy kept pressing on about this. Apparently, he didn’t get the memo [as here, but also as here.]
Which calls into question the guy’s motives and quality as recruiting-HR (didn’t find any content knowledge beyond a few job profile buzzwords). And was the reason I hadn’t too many questions anymore. An hour scheduled, now at 45mins. I found no reason to continue.
By the way, the guy himself was with the firm already for seven years. Twice mentioned during the interview that attrition rates are bad – no mention of cause analysis.
The co. is in the process of being taken over; their previous strategic strength has struggled too long now and new initiatives haven’t born fruit quickly enough. The above qua HR conduct, when (not if) translated generically on operational and tactical levels (by the guy being in his right place, or setting oneself as standard while hiring), aligns with that and then this sort of thing is out of the co.’s league altogether. Do the math [quod non].
Edited to add: This here thing is fully correct, especially the latter bullets…

Conclusion: One good, one bad, it was ugly.

Let’s close off with a positive note or two or three or…:

1. It was a tragicomic experience, from which at least I learned a bit again. Unsure the others involved, have learned or will.

2. Discussions with the chief did give me some insights into
a. where a company of this size and stature stands, qua development in our fields (plural since it’s/they’re many) – apparently, what I normally blog about is far off for the ‘GRC’ environment I was interviewing for;
b. the reason the company will get taken over hence made me aware of the need to next time learn more about such strategic developments beforehand and during interviews;
c. what my own levels of knowledge/experience and near-future strengths and wants are.

3. The external liaison called early the next morning. Attentive.

And now for something completely different:

An update. Since they wanted a second interview, to gauge the fit of character.
This time, with a peer – that frankly did not display an understanding of the subject matter to be riskmanaged beyond much generalities – and a staff; enthousiastic but maybe not really with the long track record and education to fathom the finer details of the subject matter or history of it, let alone latest (last-decade) developments in it.
Yet another time-boxed meeting, with all sorts of questions that corroborated the above, no need to repeat at length. These two seemed like decent people, in need of support indeed. Note-taking, didn’t happen too much on either side. No, I did not give the rundown on what steps to take in situation A in case of B, one by one. The interview was to be about fit of character, not an instruction course. Yes, I did talk about how I would handle certain management issues.

Bringing up the subject of ML in trading at an industry-related co., the peer dismissed that as “Oh yeah, that’s merely implementing trade strategies”. Rrright; not strategy but execution, and not strategy but arbitrage hunting. And not done at this firm at all; at a tone to make clear that AI is fringe to the financial industry.
That company caters to, one mus be frank not the higher tiers of private investors [as core, vast majority part of the business – they have tried to diversify but had to admit in the takeover paperwork they had not made much progress with it which given all the above may not be overly surprising and giving a reason for being a takeover target maybe?], but shouldn’t they be dabbling with it as a service to their customers …?

Next, we discussed the takeover by the foreign co. … Yes very busy reorganisation times are ahead [in all areas but not the GRC one it seemed to be implied; how did they manage to give you that idea ..??], but hey this firm is a perfect geo fit to the Other’s portfolio and by the way they are in wholesale and we are in retail so it is obvious that we will remain a separate entity all the way. Yes of course that’s why a. you are all prepped to be very busy – with what, you think? you already stated to be busy and b. the official takeover paperwork states that your platform will be out, and theirs in. [I read that paperwork, but couldn’t notice that you had, too.] Your platform that serves still the vast majority of your business – and that new one will be managed and run from HQ that is in another country. So of course your co. will not change that much ..?? ‘It has already been agreed that the local name will stay’ [as is in the takeover paperwork too]… Right again, as always happens in the history of takeovers. Always, in a full 100% Not fashion. Three years maybe, then two years of having a ‘A XYZ company’ as tagline, and then swoosh, history. Or, if the Other finds out about management and systems quality in more detail, earlier when more than the token manager is replaced…

Yet again, the impression that ones were very busy with ‘being in GRC/3LoD oh how important and core business we are’, crowding out actual business development for the good with Framework Muss Sein.

Awaited their reply. Felt awkward that if their reply would be about meeting over terms & conditions, I would turn them down just like that. The salary would most probably not be more than about 70% of what I interviewed for some time before (where they thought me somewhat overqualified, at a seriously more difficult and unstructured environment…) which would not have been a problem if, big if, the job would be interesting enough. But the latter ‘compensation’ I probed to be not.
Got feedback a week later, via the external liaison again, nothing from them directly – they were, uhm, looking for someone more operational. As if I hadn’t guessed from the questions, same as here. My (close-to-audit) opinion: this.

Sigh and bye-bye.

And:

[The guy was from this era, now in ours of AI-driven flash auto-trades…; Valencia]

Which you may interpret in the dystopian way as the expected future: Humans being slaves to robots. As long as robots have need for us. Then, we’re superfluous [duh] and environment-unfriendly, to be entertained as pets or zoo animals at best. Until that, too, runs out of fashion.

But hey, ‘robots’..? Why would we have something like physical populations of them, when we better talk of ‘systems’ – morphing into one big post-singularity System..? DARPA’s doings may be a smoke screen. Like described here, and here, and here.

Or… a strand/thread not often explored I think, think Evolution: Isn’t it that any small ‘improvements’ are made to the situation as-is, thus barring any Pareto-optimal switches and resulting in the new optimum adaptation being based on the best available of today? This could mean that if, big if, we are able to bend progress to our ways, and are not pre-emptively passed by / outmanoeuvred before we realise the seriousness of the game [?], we stand a chance. Maybe not a really good fighting one, but still.

In that case, this serious piece may be helpful. Us keeping robots as slaves. However overly short and misrepresenting that summary, I hope it entices you to study the linked.

Or get loosened-up-minded by studying this masterpiece.

For now:

[Contemplate; Notre Dame by roof colour, but what city ..?]

“A human must turn information into intelligence or knowledge.” — Grace Hopper

Indeed, too much of the world these days seems to run on information. At most. We blogged about this, earlier, when We were even more optimistic.

No wonder all the craze is to get more data. That isn’t the new oil but you had heard that, didn’t you? When not in depth, when not being able to move up, one moves sideways at best, trying desperate to get higher but slipping, sliding, failing. So many data lakes, seas, oceans; so little bare intelligence let alone Higher.

Can we fix this ..? Is there hope ..? Well, there’s this diamond.

[Amazing maze grace space; Mezquita of Córdoba]

I’ve been known to have opined, sometimes in words that have not always been the most diplomatic euphemisms, on the idea of Value in e.g., goal setting for IT-departments. To name just one example, where those that should know better, want ROI but for the R have nothing but ‘Value’, that’s what you deliver when you deliver what The Business wants.

Now, “As the business innovator W. Edwards Deming taught, if management sets only quantitative targets and makes people’s jobs depend on meeting them, “They will likely meet the targets — even if they have to destroy the enterprise to do it.” In Deming’s estimation, more than 90 percent of the conditions that affect a company’s performance can’t be easily tracked and measured. Yet managers spend more than 90 percent of their time monitoring and analyzing some form of measures; most notably, time tracking” (here).
Also, “The perception that the world is quantitative and that business is therefore mechanistic has for the past fifty years shaped all the variants of strategic planning, financial analysis, budgeting, cost management, and management accounting that have been taught by graduate business schools and practiced in large organizations. Executives versed in such practices and who believe that reality is defined by quantitative measurements are like the puppies who believe that the fence defines reality” (same).

What can be measured, is of no value; what is of value, cannot be measured.

Yeah I know a few, a precious few circumstances where Value is in the measurable, and there, sometimes the milliseconds are the long run. Short run is shorter, indeed… But for most, ditch the idea that you know what Value is about. Re-engineer your organisation indeed [even if that might be painful as you’ll probably have to transform irrecognisably], and conform to the original theory of firm. Then, durable success may be yours. The times, they are a’changing ..!

Cheers, and:

[Art, as a prime example; Barça]

When dealing with complex problems…

Since you can call anything you don’t understand complex, but most probably are just over your head and can’t even see the issue is just complicated but no more than that – being more than you can handle already but even that you don’t see. I.e.,

Complicated is a problem that needs a complicated algorithm to solve. Mostly by dumbing down the situation, a.k.a. ‘modelling’. Which is OK if the problem at hand is complicated indeed, meaning that a step-wise analysis/detailing actually works and one can prune the hierarchical problem tree of insignificant branches until one arrives at a model which can be understood (by you) and tackled with a known problem-solving strategy/algo. Most people tend to just strip down a problem’s structure by slashing away at its parts in- or too-little-discriminately as they don’t see what they loose in the process or couldn’t care they just need a model they can understand not a model that is appropriate for problem resolution – leading to Dunning-Kruger extremities. You know, like ‘any crippled model is better than no model at all’ in this way.
< Failure Level: Maximum+ >

Complex is where this all doesn’t help. Even when you do an intelligent complication stripping … you end up with interrelations tat are just too much to cope, conceptually, and no algorithm for (linear …!) problem reduction can help. Nor can any complex though process. a., we may not be capable of that to the degree necessary nor to the degree we think we’re capable of … [D-K again]; b., no progress may be made.

So, … when “Wicked problem: [A] social or cultural problem that is difficult or impossible to solve for as many as four reasons: incomplete or contradictory knowledge, the number of people and opinions involved, the large economic burden, and the interconnected nature of these problems with other problems.” – here – the latter applies here.
Even when we’re in sub-wicked, just complex territory.

To consider, when you face a ‘problem’…: If it reeks of non-triviality, go Design mode. If it truly is merely complicated – go ahead and solve it what are you waiting for.

By the way; read more on the subject here, and here again.
Shall we stop throwing ‘wicked’ around? It applies so much less often than now, and it inflates the true nature.

Now then, …:

[Great design; Rijksmuseum]

Or, should I say in typical AI style, a new linear separation ..?

The latter, not too literally. But in the whole thing, I suddenly [referring to, say, ‘over a quarter’] see new insights. As per the old “oh I have no clue let’s take an AI system as a black box and don’t want to have anything to do with the content” of ‘ethicists’, now moving over to some that have a clue and make progress. And even, from a country not known to be interested … have serious legal back-up [attempt] in that explainability thing. But I digress… back to the Clue part.

In particular, the distinction between perception and conceptualisation. Harking back to neuro/psycho/functional left-right brain differences [and how they’re connected into hopefully integrated wholes!], but also to e.g., Kant’s vision (pun intended, for the initiate) of how the brain works – not far off the percept-conceptua divide, actually – and deeper down to e.g., Zukav‘s interpretation of absolute SotA theoretical physics.
[ Note though, that claims about having to go to ‘first principles’ are the pastiche of a pop sci variant of Kant’s apriorism only; don’t trust those that often don’t progress beyond merely meekly bleating about those fp’s and never get anywhere but lost … ]

Yes, let’s work from such fundamental differences, when aligning the full scale of linear-logic algorithmic / Expert-System [ / Evolutionary as an intermission ] / Fuzzy-Logic / ML / AI to see that what we’ll want in the end is a bit of all, surf ‘n turf,
but also that as we hoomans now stand and usually have a bit of all as well (mainly to the right [zero endorsement v.v. ‘left’ or ‘wrong’] of the scale), how do we asses the efficacy of ‘our’ operations, qua ethics, correctness, efficiency ..?

We want to hold ‘AI’ accountable before allocating agency. Do we do that with humans, or is it a matter of being around and then being allowed to run for president ..? [As a pointer, not to anyone in particular, again]

Much work, very interesting:

[Still no clue how ‘they’ could’ve let Hermes be there at the entrance, but somehow relevant; Siena]

… Yes, again on auto cars [for over a century known as auto-mobiles or just “auto’s”]. This time, a possibly new subject: Who has the self-driving capabilities ..? In a sense close to ‘ownership’ that is. As training is so hard that “As a result, most companies presently guard their machine learning data as trade secrets. But trade secret protection has its drawbacks, one of which is to society at large. Unlike technology taught in patented disclosures, which could allow a new entrant to catch up (provided it licenses or designs around the patent), an AV data set of an unwilling licensor is not obtainable absent a trade secret violation or duplicating the considerable miles driven.” [1] — in a distance, it reads a bit like this. Not (only) re the Old Guard, but moreover referring to the stand-offish pose of AI gone beyond AGI… Think that one through.

In practice, for now, it leads to “Keeping AV data secret also creates a “black box” where consumers and authorities are unable to fairly and completely evaluate the proficiency/safety of the AI systems guiding the vehicles. At most, consumers will likely have to rely on publicly compiled data regarding car crashes and other reported incidents, which fail to adequately assess the underlying AI or even isolate AI as the cause (as opposed to other factors). As it is, AV developers’ “disengagement reports” — those tallying incidents where the human attendant must take over for the AI — vary widely, depending on how the developer chooses to interpret the reporting requirement. Without comparable data, consumers are often left with nothing more than anecdotal evidence as to which AV system is the safest or most advanced.” — yet again, intransparency.

Intransparency, at one great blocker of agency … allocatability, in normal human life. Here, we may have a problem..? Wrong Quote alert..! as here.
Nevertheless, this brings back the discussion about driver’s licenses. Currently, car-driver by car-driver.
Soon, both!
As the cars will have to be certified categorically, per software release per physical platform! As the interplay (and compatibility) may vary wildly, and e.g., sw may bloat and hence spoil response times even if tech capabilities don’t undo / worse the sw update(s). Every time, all the time. These failure phenomena creep, too, not only stepwise turn up. [how’zat for brevity?]
As the drivers will have to handle all sorts of exceptions when in all sorts of platforms. Yes, they‘ll need drivers’ licenses still, and be able to demonstrate much more competency, experience and skill in much more of the exceptional circumstances in which they might have to take over from the autos. Driving round the block is useless; shitstorm/clusterfuck traffic is what you get your license for in the first place! So many people will fail, so many will never be able to get the experience (from where, would you suggest ..?).

So, agency … Plus:

[Ah, and there’s NY traffic …]

[1] Quoted from this great introduction.

Yup, another one, actually from the same series: A useful reading on the tribulations of fake news, in a way. In another one [way], the scope is much wider. Last Friday’s Friday’s So(m)bering already referred to Ortega y Gasset. This one, too.

Again; how do we fight these things …? One human at a time; if only we knew what ‘time’ is … Even Kant couldn’t make much of it [re: Kritik der reinen Vernunft – no really, all that it has is a derivative notion not an operational definition] – anyways that takes too long.

OK then, leaving you with: