Struggles at the Desktop

[Warning: This article may not be suitable for readers of a sensitive disposition. It describes encounters with information technology that may be disturbing to some.]

“Nowadays if there is an error in the input program the computer not only detects it but gives the approximate description and location of the error and recommends procedure for correction.” (Gerald S. Hawkins, in Stonehenge Decoded, 1965)

When IBM hired me as a trainee Systems Engineer in 1969, it was not because of my data processing skills. That day in late August, when I walked into the Katherine Street office in Croydon, Surrey (shortly before the branch moved into the new building on Cherry Orchard Road), I did not know the difference between a punched-card and a paper-clip. It was not a classical career beginning, and not a carefully-planned strategic move. In an indecisive third year at Oxford, I had applied to take the Certificate of Education after the completion of my degree in Modern Languages, but soon began to have doubts. On a weeklong visit to a local primary school in Purley, Surrey, before the first term started, I had innocently queried the headmaster as to why the classes did not appear to be learning multiplication tables by rote. “Oh, Mr Percy!”, he replied with a condescending smile. ‘We don’t do tables any more!” For this was the era of ‘child-centred’ learning, where every infant had to discover for him- or her-self that 7 x 8 resulted in 56, and so on. I recall the way that tables and mental arithmetic were drilled into my generation about fifteen years earlier, and how the pattern of number combinations has stayed with me ever since. In 1968, however, I was entering the world of Progressive Education.

Perhaps my aspirations were also checked by my term of teaching-practice. Having had a term of almost total inactivity, owing to my being on crutches because of a rugby injury, I was informed, in December 1968, that I was urgently needed as a replacement at Bognor Regis * Comprehensive School, as the previous teacher of Russian had been fired for getting one of his pupils pregnant. I did not learn the cause of the summons until I arrived: the school was also going through a painful merger of a grammar-school with a secondary modern, which also dampened what remained of my enthusiasm. Halfway through this term, I decided that a quick return to the classroom was perhaps not the most life-enhancing prospect to be contemplated. Taking advice from some outfit that suggested that my interest in chess, bridge, crosswords and logic puzzles might open up some doors in the computing industry, I secured interviews with some manufacturers, of which NCR and IBM were the most satisfactory. I took care to complete my Certificate of Education so that I could have a back-up career lest the corridors of business found my talents wanting.

[* Bognor Regis is a coastal town in West Sussex. It gained its regal addendum after King George V recuperated there, and the monarch’s dying words have been apocryphally reported as ‘Bugger Bognor!’. When Ursula Kuczynski (agent SONIA) needed a place for her children to stay while she returned to the Soviet Union in the summer of 1938, she left them with a friend in Felpham, which is part of Bognor. There is no truth to the rumour that I was in 1969 undertaking, under deep cover, some early sleuthing into Sonia’s contacts.]

Unfortunately, IBM was a little slow to snap up the opportunity to make me an offer, so I had to write to them to explain to them that this entrepreneurial youth was thirsting to make his contribution to the computing revolution. Perhaps the company was waiting for such a show of initiative, since I was rewarded with an appointment at the Head Office in Chiswick, to meet one of their Personnel Managers (no ‘Human Resources’ in those days: employees were certainly not ‘associates’, and customers were assuredly not ‘guests’). I was delighted to find that this benevolent soul had also studied Russian at Oxford. He started to quote me a quatrain of Pushkin’s, which I was happily able to complete. I passed the interview. I was in.

Before I started the eight-week basic training course at Sudbury, Middlesex, I had a week in the office, where I was directed to a small room, and given a Programmed Instruction text on IBM’s System/360 to work through. These matters were all rather daunting to me, and I recall I had to interrupt my study to ask the Systems Engineering Manager what the meaning of some concept was. It all comes back to me quite clearly: I wanted to know what was special about the sixteen ‘registers’ of any 360 computer system. Registers were (and no doubt still are) the mechanism by which the locations of computer memory were addressed, but they also seemed to have some properties that lent themselves to high-speed arithmetic. Somewhat confused, I asked the manager whether he could explain their nature to me. “Oh, I never really understood all that stuff”, he said. “I wouldn’t worry about it.” I think we adjourned to the squash court soon after that, and I gave him a good runaround in return.

The Systems Engineering class was tough. All new recruits were required to go through the same basic training, to make sure they were immersed into the IBM way of doing things. I recall a few students who had already served several years with IBM’s rival, ICL, and were thus already very familiar with the concepts and practice of data processing. Most of the graduates straight from university had scientific backgrounds, and had used computers in their laboratory work. There were times when I wondered whether I would make it. My ability to learn seemed to correlate exactly with the ability of the individual instructor to present topics in schemas that matched how my brain was able to integrate new ideas, namely very logically, with clear step-by-step evolution, and no grand jumps that left canyons of unexplored territory behind. Gradually, things began to make sense. I completed the three stages of the training about a year later, and was ready to roll.

Unfortunately, IBM was not sure at that time exactly what the role of systems engineers was, as anti-trust threats had meant that the company could not hand out systems engineering resources to its customers for free. At the same time, we were neophytes eager to learn by practical experience, while the projects we were given were haphazard, not always suitable, and not always educational. I soon learned that I liked coding, appreciated the value of well-designed and well-implemented systems, and became very frustrated with poorly written documentation. And I did have a knack for working out what was at fault when things went wrong, although that experience was marred by a disastrous project where I was asked to make some changes to a Vehicle Scheduling Package for a prominent and demanding customer. There was no guide to how the product worked, and I stumbled for weeks in trying to tweak it to meet the idiosyncratic needs of the customer. I received no help: the project was simply abandoned, I believe. But two lessons started to emerge in my mind: i) the knowledge that there was a logical explanation for every computer failure, and ii) the importance of good diagnostics being built into any product.

I move forward seven or eight years, and two jobs later. I was working as European Customer Service Manager for a small American software company. Our flagship product was known as a transaction-processing monitor, an adjunct to the operating system that handled communications with a network of terminals and managed the user programs that the customer wrote to provide on-line business functions. One of the challenges with this software configuration was that a motley set of technologies all operated in one partition, all clamoring for resources, and all potentially stepping on each other. Much of the code was written in low-level Assembler language, which provided greater manipulative power, and faster execution speeds, but also provided opportunities for corrupting storage occupied by other software. Frequent were the ‘core dumps’ (we still called them such, even though ferrite cores had been superseded as memory components by then) that were mailed in by customers when the system blew up, and we were unable to detect what had happened over the telephone. Then the support team would compare the state of computer memory with source listings of our product, in order to find out where our product (it was frequently the fault of the product) had gone wrong.

One particularly stubborn problem endures in my memory. A prestigious customer had experienced an execution failure, not recreatable, that caused the partition to explode. (The customer was actually the institution where the atom spy Klaus Fuchs, for whom Sonia had acted as courier in 1942-43, was working when he confessed in January 1950: there is no truth in the rumour that I was sent in by MI5, under subterfuge, to undertake an audit of its security procedures.) The requisite hundreds of sheets of print-out were sent in. No one could work out what had happened, and I devoted hours and hours to it. Eventually, I determined that it looked as if an error routine in IBM’s telecommunications package, VTAM, had failed to save properly the register contents that had been passed to it (and which had to be restored when the routine had completed its work), as all processes used those same registers I had been puzzled about back in 1969. I called the customer with my tentative suggestion, and asked him to pursue the matter with IBM. The next day he called back: indeed, one of the error handlers was incorrectly saving and restoring registers. He apologized for not searching for, and applying, the fix that would remedy the problem. Much goodwill was gained.

The second experience that reinforced my earlier lesson was in helping to roll out a new feature in the product, something called ‘Multiple Record Hold’ (MRH). The previous version had allowed only one file record to be held at a time, which was a heavy constraint. If a user application wanted to prevent anyone else accessing a customer record, say, while it then checked an inventory record that it might want to deplete, the systems designer was in a bind. MRH addressed that problem. But our developers designed and coded the feature too quickly and carelessly. Several occasions would arise where the programmer would try to invoke the feature inappropriately (for example, invalid keys, or multiple requests to the same file), or the software detected something illogical. It would return an ‘L’ code to the program, indicating such. But the programmer had no idea what it meant. There must have been several dozen places in the source code where an ‘L’ error code was returned. We, as support personnel, had to trace through the record of programmatic requests, and the source code listed, to detect at what point in the logic the ‘L’ had been returned, and then provide an explanation. But it could all have been made so simple: an auxiliary area existed where a return code could have been posted, and a corresponding piece of documentation could have explained what every code meant, with an enormous benefit in productivity. I was just about to start coding this enhancement when I was invited to work as Director of Technical Services for the parent company in Norwalk, Connecticut. At that time the flagship product was on the way out: the feature was never implemented. And so my wife and I, with ten-month-old son, moved to the USA.

[In parenthesis, for the more technical among my readership, I should also mention here that an unusual feature of this product was that the Control Program was written in a high-level language, COBOL, a decision presumably made in the interests of clarity and maintainability, not in the cause of performance. But when some advanced features were added to the product, it became necessary for the CP [not the Communist Party] to access low-level bitstrings, something COBOL cannot do. Thus an Assembler (low-level) language subroutine called GETBITS was added, to return statuses for further decision-taking and logic-branching. I recall very clearly how one of our most demanding – and shrewdest – customers in the UK, when undergoing performance problems, ascertained, through the use of a testing device, that GETBITS was consuming 6% of all machine cycles on its 370/145 – an enormous amount. Furthermore, when I inspected the new CP code, I discovered that, in many circumstances, the GETBITS routine was being invoked, but the CP was then taking branches that were completely independent of the results of the call! When I vaguely suggested to the President of the Company (who had probably written much of the original code himself) that I could rewrite the whole CP in Assembler language on my weekends, and deliver a much faster system to our customers, he declared, very seriously, that anyone who attempted that would be fired. He still relocated me to report to him in Connecticut, but later gracelessly told me that he only did so because the Director of R & D persuaded him. On such whims do whole lives change.]

The reason for this long introduction is that I recently had to replace my home PC, and experienced massive problems. For some months, my old HP Pavilion had been warning me of its imminent demise. The fan had broken, and the device was presumably in danger of overheating. I would get a warning message each time I re-booted, and occasionally Windows would blow up. So shortly before Christmas, I bought an HP Envy Desktop, preparing to install it after my winter break. I did not buy a printer or monitor: I had an HP Photosmart printer that was working well, and, only a year ago, I had had to replace the monitor that had suddenly died on me with a new model. This new monitor had HDMI support, but, since my PC was so old, it did not support an HDMI connection, and I thus had to use the older-generation VGA connector. This apparently meant that I had no sound support on my computer, but that was no great loss, even though I could not listen to music while I was working. I got used to it. Early in January, I thus loaded up the printer with new ink cartridges, backed up the files on the old PC, checked the cable configurations to ensure I knew what socket went in where, and unpacked the new machine.

To start with, all went very smoothly. True, Windows10 was a bit of a shock, with some features apparently dropped, and some weird patterns of activity occurring, such as random duplication of keyboard strokes. But overall it worked, and I restored my files (well, partially: but that’s another story.) Then I suddenly realised that I was not getting any sound from the computer, despite the new HDMI connection. The driver was okay, the system told me that the graphics was working properly, and yet no sound emanated from the monitor. It took me a while to work out that, all that long year ago, I had been sold a monitor with no sound support. Well, it was my fault for not asking, I suppose, but I think the salesperson was at fault, as well. Maybe he just wanted to move that product off the shelf. After all, why would I want to move from an antiquated broken monitor that supported sound to a spiffy new one that didn’t? There’s a lesson.

Next, I tried the printer. And here is where the problems started. Word documents would not print at all; PDFs would print, but very faintly. Crossword grids from the Web printed out partially. Emails from my queue printed fine, however. (I have one to prove it, as it relates to my problems.) What was going on with my device, which had been working so well a day beforehand? What caused such erratic behavior, where some items came out fine, but others were ignored? I did not believe it was a dirty printhead problem (something I had encountered and fixed a couple of years ago). My first step was, on my next trip into Wilmington (thirty-five miles away) to go to Best Buy, the store where I had bought both the printer and the PC, and ask for their advice. They immediately said ‘buy a new printer’, hinting that many users suffered from the same or similar experiences, as it would be too expensive to investigate the problem, and printers were so cheap. But I wasn’t going to give up that quickly.

After looking on the Web for users with similar complaints, I tried a number of things. I reloaded the printer driver (the current version was dated October 2015, which was perhaps not encouraging). I deleted the device, and added it back in. I reset it. I set it up as a default printer. I tried printing test pages. At some stage I logged on to the Microsoft and HP support forums, where ‘experts’ (but not employees of the respective companies) would generously offer suggestions to fix the problem. Nothing worked. Eventually, an HP employee joined the forum, and tried to help me. I shan’t go through all the steps he recommended, but he ended up giving me secret codes to enter on the printer itself, to determine why it wasn’t able to operate any off-line functions either. But even this process did not work as he outlined, as it was interrupted by another message. At this stage, we agreed that I should call up HP customer support.

Since the problem appeared to be with my newly warranted PC, I called the number for desktop computers, and was soon speaking to a support representative (in India), to whom I gave all the relevant information. Then, when I described my problem, he said that I needed to speak to the Ink-jet support group, and gave me another number to call. I went through the same process, was given a case-number, and started providing details of my problem. But when I gave the representative the Serial ID of my printer, she (in the US, this time) told me that I would have to pay for support, as the device was no longer under warranty. This did not completely surprise me – I have paid for such telephone support from HP beforehand – but I was not actually in the mood, given the trials I had already experienced, for having to pay for diagnosis that I really felt was HP’s responsibility. I somehow convinced her that she should at least provide an initial investigation of the problem for free. So we downloaded some software that allowed her to control my computer while I watched.

What happened next was rather disturbing. The representative asked me what make of router I was using, and when I responded ‘Ubee’, she expressed a degree of shock, almost one of recognition, as if the Ubee-Photosmart combination was a known toxic one. I tried to determine whether that was the case, but received no reply, as she started manipulating the Ubee tables on my PC. Clearly, she knows what she is doing, I said to myself. And then the connections were lost. First, the phone contact disappeared. She sent me a message indicating such, so I quickly sent her a text, imploring her to call me back. Then that connection went dead, too, and I was left stranded, with the shape of my router tables unknown, and the problem unresolved.

At least I had a case number. I called back, but this time was routed to another call-centre in India. Even though I gave the representative there the case-number, and told him what had happened, he claimed he could do nothing for me. I rung off in exasperation, hoping that the contact in the USA would call me back. But nothing happened. I suspect that the supervisor of the representative trying to help me in the USA had interrupted the process, probably reprimanding the young lady for not charging me for such support time, and thus had broken off all contact. I shall never know. Even when an HP customer relations person (who had presumably kept an eye on the forum, and had been alerted by the HP technician who joined it) contacted me afterwards, he was powerless to find out what had been going on. But to abandon a customer half-way through a process when the device was under the control of a remote technician was scandalous, in my unhumble opinion.

So I gave up, and bought a new printer, from Epson. Never again any HP products for me.

Perhaps it was all a strange coincidence, but one afterthought came to me. If my printer had enough intelligence in it that, when I ran out of ink, and inserted new cartridges, it could send a message in real-time to HP Central to encourage me to buy a replacement set, maybe it was also smart enough to detect that it was now being driven by a more modern, faster computer, and that a process akin to what we systems engineers used to call ‘graceful degradation’ should occur, so that the user would have to buy a new printer? That was the immediate recommendation of the technician at the company who sold me the printer, remember. After all, Apple has admitted slowing down its devices to preserve battery power, and Volkswagen fudged emissions when engines detected that they were running under laboratory tests. I would not be at all surprised if something like that happened.

And then my wife’s laptop computer started having problems. She would be told that an important Security update needed to be installed on Windows10, after which the process would hog her computer for hours on end, only to fail with the message ‘0x800700c1‘, when it was 99% complete. We ignored it for a while, since I was mightily consumed with sorting out my own PC, but I at last got round to investigating. ‘Contact Microsoft Support’ was the guidance, so I went on-line, and was soon directed to a document titled “You receive the error message ‘Something went wrong’, when attempting to install the latest version of Windows10.” I was amazed to learn that the company offered ‘many steps that I could try’, as there were ‘many possible reasons your device may be unable to update to the latest version of Windows’. This was extraordinary. A specific error message had been issued, yet the software had no clue as to what circumstances had cause it to fail, and the user of a consumer product was supposed to experiment with all these approaches in order to resolve the problem? What on earth would the Little Old Lady from Dubuque do?

I decided to request an on-line chat with a support person. This did not take long, and I was put in touch with Parthiban, in India. We set up the protocol by which such persons take over control of the computer, and he soon decided that the problem was due to a corrupt database, and a conflict with Norton Security. He initiated the update again, but he had to sign off before the process completed, leaving me with a link that I could invoke in case of failure. I was given a case-number, and waited for an hour or so. And then the installation failed again. So the next day, I used the reinvocation, and was before long involved in another on-line chat, with Deepthi. Now Deepthi did not appear to know what he (or she) was doing, as I could watch him wandering aimlessly around HP configuration options. My mistrust was justified, as he suddenly signed off the session without letting me know why.

Accordingly, the next day, I reinvoked the link, and noticed that I was 93^rd in line, so decided to try again later. The queue had then diminished to 21, so I tried it again, and was soon engaged in an on-line exchange with Praveen. His diagnosis was that some cookies needed to be removed, and Norton Security had to be disabled for a while, as it was inhibiting the execution of the Microsoft Update routines. So I watched as he cheerfully went through the whole process leading up to the installation of the updates. Then he left me to watch for an hour, until the update failed again.

Yet, when I tried to re-invoke the link to resume my interchange, I was told that it was no longer valid. This time, I resolved to speak to a real person, called the support number, and, after a wait of about fifteen minutes, I described my problem to the support representative. She took my number, and soon I was talking to another agent, named Tony. (By asking him what time it was where he was working, I determined that he must be located somewhere in the Mid-West.) Anyway, while he seemed to be unable to look up my Call Number, and discover what approaches had already been applied, Tony sounded much more confident, and judged that I needed a larger partition size to run the routines. So I watched as he downloaded the Minitool Partition Wizard (how come Microsoft does not supply this facility?), which ran for about half an hour. That task having been successfully completed, he said he was going to re-install the whole of Windows10, so that I would not have to deal with a separate Security Update. I was getting a bit anxious as this process started, so I begged him to stay on-line until it completed, indicating that he could multitask with other customers while the update continued. Yet he was so confident that his solution would work, he said we should ring off: he did however commit to calling me in another hour to check how things were going.

Predictably, the update failed. After about an hour and a half of installation, verification, preparation and execution, I received a short message, with no diagnostic code: ‘Windows installation has failed’. And this saga would not be complete unless I informed you that, no, Agent 4 (Tony) never called me back, despite his promise. I had been abandoned again.

Before finally agreeing to give up completely, and simply to ignore the messages emanating from Microsoft that were constantly bugging my wife, as she worked at her computer, informing her that her security was at risk, and that updates still needed to be installed, I decide to post a plaintive appeal on the Microsoft Support Forum. I summarized all that had occurred, and expressed my frustration at Microsoft’s shoddy installation software, and its even more unprofessional support agents, who appeared to apply guesswork in trying to resolve problems, and repeatedly left consumers like me hanging dry. My appeal was quickly picked up by a Microsoft employee who has been very patient in going through my experiences. Yet his final recommendation, after I gave him the status of my Windows10 System Build, and maintenance applied, sounded very much like the process that Agent 4 had undertaken. When I pointed this out, he urged me to try what was (he said) a very simple process: indeed, he himself had written the on-line document that guided it. So I sat down, went through his steps, disabling Norton Security and trying again when that package told me that one of Microsoft’s modules was unsafe, and had had to be removed. About ninety minutes later, the Microsoft software, having gone through download, installation, verification, and preparation, started its execution. After half an hour, I received exactly the same message that had appeared in the previous try: ‘Windows installation has failed’.

The Forum Observer responded promptly, requesting that I send him (via OneDrive) a couple of log files from an obscure Windows folder. I am not sure why no one had thought of inspecting such data before (I had in fact suggested such a course of action several days earlier, as I suspected such files should exist somewhere). I had not used OneDrive (Microsoft’s file-sharing service on the Cloud) before, but I retrieved the logs, followed the instructions from my iPad, created the OneDrive link, and posted it on the Forum page.

And then I received the following amazing message from the moderator:

“A Windows upgrade requires DISM utility to work and in your case DISM fails which then triggers a rollback.

Error initializing DISM Session: [0x800700c1], [gle=0 x800700c1]

Right-click Start>Command Prompt (admin) and type in:

DISM /ONLINE/ CLEANUP-IMAGE/ SCANHEALTH

If that fails with 193 post back the DISM log present at C:\Windows\Logs\DISM\ again through Onedrive”

As John McEnroe would say: ‘You cannot be serious!” And don’t you just hate it when your DISM fails? So I went ahead, and yes, the SCANHEALTH failed with a 193, and I posted on the forum the link to the DISM log on OneDrive. Isn’t this exciting?

The next news was not good. My contact thought that the damage ‘was beyond repair, and that I would either have to reset Windows or do a clean install. He pointed me to another link, where a Mr Carmack had published a document titled ‘Clean Install Windows 10”. Mr Carmack attempted to sell the process by describing it as ‘a game-changing learning experience that will make you permanently the master of my PC’, going on to write that ‘to stretch this out over days or weeks you’ll learn better how each change affects performance.’ But typical home users of PCs do not have ambitions of becoming geeks, taking up Windows maintenance as a hobby. The only game I wanted changed was the one of getting Microsoft to fix its software. The steps that Mr Carmack outlined are monstrous (see https://answers.microsoft.com/en-us/windows/forum/windows_10-windows_install/clean-install-windows-10/1c426bdf-79b1-4d42-be93-17378d93e587), and must be very prone to error. And, even if I went through all this, what were the chances the problem would recur? I replied in this vein, thanking the moderator again, suggesting it was perhaps time to give up. ‘So what are the implications of simply ignoring the attempts by Windows to install the Security updates? Maybe the laptop should simply be replaced?’, I asked.

There is an easier way, replied the moderator. He outlined some other steps, recommending that I do a reset, ‘as it might remove the corrupt driver which is preventing the upgrade’.  He had no idea what might have caused the problem, and suggested yet another site ‘where the experts might be able to help you better’. But, if a driver has been identified as defective, I wondered, why could it not be replaced? At this stage, I concluded that I had had enough. My wife and I would live with whatever nonsense Microsoft imposed on us, and replace the laptop with something from Apple when the time came.

It was difficult for me to imagine that my wife’s PC was the only one on the planet undergoing such experiences. She is a woman in a million, I know, but I do not understand how her rarity should extend to the tribulations on her laptop computer. And the exercise also reminded me how little way the software industry (or Microsoft, at any rate) has come in fifty years. The company delivers an upgrade to a system that is in many ways incompatible with the previous versions, and it has disabled certain functions. The on-line documentation frequently does not match how the screens of system information appear, so one is left groping. The diagnostic codes given when the software encounters problems are meaningless and obscure. One can find jokey tutorials on YouTube, but they are badly designed, often delivered in mumbles, and do not explain enough about the Whys of a particular feature. The support personnel who try to help the bewildered consumer are poorly trained, not provided with proper tools, and thus engage in guesswork. And, of course, we fogies have to deal with tracking down those tiny labels with product serial numbers, pasted in the most inaccessible places on the equipment, that have to be read with a magnifying-glass.

What galls me even more is that we (in the USA, anyway) are currently facing a bombardment of in-your-face advertising from Microsoft that promotes its new expertise in Artificial Intelligence as ‘Empowering Imagination’. It depresses me to think how such technology will be abused by a company so obviously inept at managing the release and maintenance of its own software. Perhaps the techniques of neural networks should be applied to Microsoft’s own configuration and diagnostic problems before they are imposed upon an unsuspecting world? Yet again, we have been here beforehand. I recall the surge of enthusiasm about AI about thirty years ago, when all number of hyperbolic claims were made about the advent of rule-based systems. Now we hear it again, with all sort of nonsense about systems that will be able to teach themselves how to be more effective, and thus achieve all manner of breakthroughs in medical diagnosis, or fraud detection, or whatever. Computers can be programmed to give results that appear to reflect intelligence, such as beating grandmasters at chess, but that does not mean that they are inherently intelligent.

Maybe this generation of AI is different, but a caveat remains. A key principle of computing science has been the verifiability of systems – the fact that code must be inspected to determine whether the logic has been implemented according to specifications. (If proper specifications actually exist, of course, which is a whole other problem: see Multiple Record Hold.) Thus I used to experience the process of ‘structured walk-throughs’, where one’s peers would wade laboriously through the code a colleague had written to apply more stringent tests that might escape the test data environment. If the onus of decision-making has now been delegated to the computing system itself, who now takes responsibility when something goes wrong? I was both amused and perturbed to read, in the New York Times, earlier this month, how engineers at Google have started analyzing how computers using neural networks reach the conclusions they do, as if the experts are concerned about the level of auditability that these systems provide. “Understanding how these systems work will become more important as they make decisions, like who gets a job and how a self-driving car responds to emergencies”, the article declared. (I write this the day after the Uber self-driving car in Tempe killed a pedestrian during a test-run.) Their concerns are appropriate: I smell litigation over unexplained, and inexplicable, disasters. The paradox is that, if the processes of AI are verifiable, the technology is considered mundane and unimaginative, while, if they are not, it is uncontrollable and dangerous.  What do you think, HAL?

*          *          *          *          *          *          *          *          *          *          *          *

A few years ago, the Times of London informed me it could no longer issue a cheque for the occasional fees for published Listener crossword puzzles without my submitting a complex form that confirmed that I was a proper US-resident tax-payer. The cost to complete the forms required was almost as much as the crossword fee, so I didn’t bother. Last year, my bank in the UK (with whom I have had an account since 1965) told me that I would have to change my deposit account into a long-term instrument that would mature in three years, as it was no longer allowed to pay interest on accounts to overseas customers. This month, I received a letter from Barclaycard (with whom I have had a sterling credit card for about forty years) advising me that my account would have to be closed in early April unless I could provide proof of a residential address in the United Kingdom. Thus another convenience (for paying magazine subscriptions, downloading files from the National Archives, purchasing gifts, even ordering a copy of my own book from amazon to send to a reviewer – all in sterling) disappears. I have maintained my UK citizenship, have paid all tax at source, as appropriate, and have always declared all my (puny) UK-based income to the US Internal Revenue Service. It is comforting to know that the British authorities are cracking down on the real risks to currency and tax fraud, and thus discouraging me from any further investments or expenditure in the UK, while allowing all that other soiled money from Russia and other places to be brought into London for the purposes of acquiring valuable assets and helping the economy.

This month’s Commonplace entries can be found here.