I'm a neuroscientist with a decade in fMRI and that criticism is unwarranted. Scalp based EEG is inherently noisy but if hundreds toward thousands of people are using these, there's a real chance the signal could overwhelm the noise.
What's the alternative? A bunch of lab-based studies? Those have their own severe problems. Research isn't zero sum. Every little bit counts.
In order for an EEG signal to overcome noise it is typically time locked to an event and recorded multiple times. The signal is then averaged time locked to the stimuli. How is this done cycling around NYC? Do you politely ask a driver to swerve into your path 5 or 6 more times in order to get a proper sample of the event?
If you have experience in fMRI consider that this EEG is going to look like continual gradient and line noise artefacts, potentially highly correlated to the type of road surface the bike is currently on. Please explain how this can be 'averaged out' in order to obtain a signal.
I'm also work in (applied) cognitive neuroscience and disagree with respect to the utility of these studies (ie "every little bit counts"). Pop-neuroscience articles that make large claims have the potential to poison the well in terms of the public's interest in, and the reputation of, neuroscience. Too much hype in a field has the potential to create a backlash, as in the "AI Winter" of the 1980s.
I've watched fMRI (the God region of the brain!) studies poison the well. Pop-science is an endless, multimedia beast. Research that replicates is truth winning out. Everything else, including many high profile journal articles, is just noise.
What's the alternative? A bunch of lab-based studies? Those have their own severe problems. Research isn't zero sum. Every little bit counts.