Outing Robots

As I mentioned in a previous article, I don't like FT8 robots. I don't dislike them on principle or because they're new and less "human", but for practical reasons: they call others when they shouldn't and they consume valuable spectrum. I steer clear of robots when I can positively identify them. I might a robot just to silence it, and then forget to log the QSO.

Identifying a robot is not as easy as you might believe. Human operators can exhibit behaviour almost indistinguishable from robots, robot algorithms are frequently changed, or the operator alternates between live and robot operation. Although I'm not obsessed with robots, identifying them is useful for avoiding them. But how?

For illustrative purposes consider this fictional conversation with what may, or may not be a robot:

Human: Are you a robot?
Other: No!

H: You call CQ for hours on end, and only a machine would do that.
O: I like making lots of contacts and earning awards.

H: When you call stations you annoy them by always transmitting on their frequency.
O: That's because they choose frequencies that are free of QRM at their end.

H: You also answer a CQ from every station you haven't work before.
O: They send CQ because they want others to call them. Nothing's wrong with that.

H: Even when they call CQ DX or CQ JA?
O: That may be what they prefer but often they will answer me.

H: But you call them 20 or more times in a row.
O: A lot of stations don't hear well so I have to keep trying.

This is deliberately written to mimic the most common form of the Turing test, where a interrogator interviews a subject via text messages and has to determine whether the subject is human or an AI (artificial intelligence). It is a difficult challenge and, to be blunt, most people are easily fooled. Many humans already have difficulty correctly identifying text responses from a LLM (large language model), which is not intelligent.

The question remains: how can an FT8 robot be identified? There are three major areas of inquiry:

• Self identification
  
• Behaviour pattern
• Stimulus-response

Self identification

Robots sometimes identify themselves. This might be surprising unless you know where to look. For example, digital stations connected to PSK Reporter will dutifully identify the software application. Bring up a map in PSK Reporter and hover the mouse over the location marker over the suspected robot station (you can narrow the search by noting its grid square).

Unfortunately this no longer works as well as it once did. I tried this for many likely robots while writing this article and I didn't find one. Robot operators have learned the hard way that they need to be less obvious because some hams took to tracking them down and telling them what they thought of what they were doing.

There are two ways to mask the software. One is not to not connect to PSK Reporter. The second is to use a feature of most (all?) of these applications to misidentify as WSJT-X. For this latter method it is possible to root out likely robots in a subset of cases. Look at the software version being reported. Negligent robot operators may not notice that the version being reported raises suspicions.

I located one station connected to PSK Reporter that by its longtime behaviour is almost certainly a robot most of the time its on 6 meters. I redacted the call sign and grid in case I am wrong. Is there anyone still using this ancient version of WSJT-X? A scan of other connected stations on the map will soon convince you that the use of this ancient software is rare.

Behaviour pattern

Since I discussed robot behaviour in the previous article I don't need to say much more. I would only caution robot hunters that behaviour patterns can be quite complex depending on the operator's configuration of the software. In many cases they keep it simple, opting to CQ or respond to CQs using mostly default parameter choices. Those are the easiest to spot, especially when they endlessly call CQ.

Don't be too certain that endless CQs or answering every CQ are sure signs of a robot. Human operators do that too. Cancel the watchdog timer and you can call CQ forever, and that's what some hams do. They use the auto-answer features of WSJT-X and JTDX and glance at the monitor from time to time in case there is a QSO in progress. When it's done they log the contact (or make logging automatic) then manually re-enable the transmitter to resume CQing.

Quite a few human operators pounce on any station they haven't worked before (those CQs have a distinct colour code). When the contact is logged they pounce on another. Most hams are more discriminating about who they call, so it may see robotic to see stations that call anyone and everyone they haven't worked before.

Stimulus-response

If you suspect a station is a robot it is possible to give it a poke and see what happens. This is analogous to the Turing test described earlier. The best way might be to call CQ with an unusual call sign that almost every human operator would react to differently than most robots. 

Before I continue, I must caution you that to do this deliberately can be unethical and even counter to the regulations for your country. I've never done it but I have seen it done inadvertently. When it occurs the results can be quite educational. I will give you an example that I recently encountered.

There was a station in the Caribbean that was very popular for DXers. Many common countries are often not so common or easy to work on 6 meters. The station in this instance was VP2MKP, and it was a new one for me when I worked him. 

Apparently he wanted to work stations faster so he composed a couple of custom messages to announce that he would move to FT4 on 50.318 MHz. This is a good idea and I wish more stations would do it. When conditions are good, doubling the QSO rate is worth a few decibels of sensitivity. But let's put that aside since it is not the point of this article.

There were two free form messages that were each transmitted several times. There is a strict limit to the length of these messages so they are often somewhat cryptic.

• CQ VP2M/FT4
• CQ VP2M/50318
  

The following screenshot captures only part of what ensued. It is enough to help you to see how different reactions of humans and robots. Call signs of suspected robots have been redacted, but not their grid squares.

The robots instantly reacted to the CQ message that appeared to contain a portable indicator for what was interpretted as a call sign. Although not real call signs, WSJT-X (at my station) tagged the messages as valid CQs. The port of the same software used by the robots did the same. Obviously the robots have never worked these call signs and so they went to work.

Would a human have clicked on the messages and answered, and do so in the very next period? That is highly unlikely, and it appears that none did in this instance. I can say that I stared at the screen for several seconds until I understood what those blue labelled messages from VP2MKP signified, and there is no way I would have replied, instantly or later, since that would have been foolish.

Not only did the suspected robots call, and call instantly, most stopped calling at almost exactly the same time! A lot of robot operators clearly stick with the default software parameters. There were variations. For robots that were in QSO at the time there was a 30 to 60 second delay before calling the false CQs.

How would you have reacted at the time. I can tell you how I reacted: I laughed. Then I scrolled back and forth and to confirm what had occurred. I knew I had a great idea for a blog article so I took the above screenshot.

There are many ways to deliberately provoke robots without waiting for chance to deliver what I've shown above. I'm sure most readers can come up with at least one. However, almost all are as unethical as the robots themselves since they require the transmission of deceptive messages. I won't do it and I would discourage you from trying it. Two wrongs don't make a right.

What should you do?

I don't know. What do you think you should do? If you really hate robots, my advice is to proceed with caution. Without walking in to the suspect's shack and inspecting their station, you cannot know. Never assume that you have found a robot with any of the methods discussed here or elsewhere. Being 95% certain is not 100%. Everyone deserves the benefit of the doubt.

In this and the previous article, I hinted at how I typically deal with robots. They are an annoyance but not a capital offense. I will not allow the presumed faults of others to become my obsession. If you are emotionally incapable of that, don't expect help from any regulatory authority, club or operating award sponsor. Avoid confrontation! That also applies even for (what you believe is) a friendly approach.

Most robot operators give it up after a while. Watching a machine make contacts is ultimately pointless and boring. Of course, others will come along and give it a try but they, too, will soon stop. In my opinion it isn't worth worrying about. I'll now give it a rest and not soon talk again about robots on the blog.