This is the first of several pieces in response to questions I have received about my recent lengthy article (too lengthy!) on “Ghosts, models and meaning: rethinking the role of communication in science.” It’s intended to give a quick overview of the main ideas; you’ll find the full article here.

Can you give me a succinct definition of the “ghosts” you’re talking about?

There are a lot of contexts in which science communication somehow fails because an audience doesn’t get the point or understand a message the way it was intended. The naïve view of this is that scientists just know a lot more about a specialized topic than people from other fields or the public. Of course that happens, but I’ve found it’s rarely the biggest issue in communication. And it doesn’t explain why people so often have problems writing for experts in their own field, or have trouble clearly expressing things they know very well.

When I began teaching scientists to write, I constantly came across content-related breakdowns that were hard to understand. This got so frustrating that I finally decided to carry out a systematic analysis of the problems. That took about four years, and “ghosts” emerged as a fundamental concept that’s helpful in understanding a lot of what goes wrong.

Ghosts originate from many things: concepts, frameworks, logical sequences, various patterns of linking ideas, theories, images and so on. What unifies them is that the author has something in mind that is essential to understanding what he means – but it’s missing or very hard to find within the message itself. Often the author is not even aware he’s thinking of something a certain way. Since it’s nowhere to be found in the message, it’s invisible. If the reader doesn’t sense its presence and go looking for it, or has too much trouble digging it out, he will probably misunderstand what the author really meant. All the words might make sense, but there’s some core idea that’s still missing.

I call these things “ghosts” because they are invisible, in that sense, and yet highly disruptive. Of course they occur in all kinds of communication. But ghosts are particularly interesting in science because it has very structured and special ways of assigning meaning to things. What things mean depends on a hidden code that most scientists eventually absorb and imitate, but a failure to recognize its existence causes all kinds of problems. A scientific text will be completely opaque to a lot of people not only because its meaning depends on all of these invisible things – even more because people don’t know where to look for it, or that it’s there at all. It makes science harder to communicate and much harder to learn.

What this boils down to is that science has special ways of assigning meaning to things that really need to be taken into account when you’re planning a message or trying to interpret one. If you don’t, a lot of misunderstandings become almost inevitable, when they could easily have been avoided.

You mention models again and again – why are they so central to misunderstanding science?

Among the most significant and disruptive ghosts in science are various models that are used in formulating a question or hypothesis and interpreting the results. Most studies engage many types and levels of models. In a single paper an author often draws on basic concepts such as the structure, organization and composition of cells, to the components and behavior of biochemical signaling pathways, to complex processes such as gene regulation, to notions like states of health and disease, evolutionary theory and so on. The way scientists describe fairly simple things usually draws on a complex, interlinked universe of models that goes from the smallest level of chemical interactions to mechanisms, organisms, species, and their evolutionary relationships.

Scientists obviously recognize this; as Theodore Dobzhansky said, “Nothing in biology makes sense except in the light of evolution.” But there is a big difference between vaguely acknowledging this and actually working out how the vast theoretical framework of evolution reaches into every single event you’re studying, and reaches into the way you understand the “simplest” things – such as the names of molecules.

And often people don’t realize that even Dobzhansky’s statement is resting on huge, invisible ghosts that he doesn’t explicitly state but are essential to understanding what he means. What I mean is that evolution itself is based on principles of science that are even more fundamental – it follows from them. So if you’re talking about the theory, you’re also engaging this deeper level. That’s really interesting because most of the “debates” over evolution I’ve witnessed are actually arguments about these even larger things. If the parties in the dialogue never articulate that deeper level of the disagreement, it makes very little sense to discuss the types of details that people go round and around about. They’re exchanging a lot of words, but they don’t fundamentally agree on what those words mean. They are arguing about whether species change, split apart or go extinct, but to get anywhere on those issues you have to agree what the term “species” means. It’s not so much that they don’t agree – more that they don’t even realize there is a problem.

What deeper ghosts have to be faced before someone can really understand evolution? 

I think there are two, which are so basic that they distinguish science from other ways of thinking about things and assigning them meaning. I call the first one the principle of local interactions, which follows from a fundamental assumption about physical laws. In science if you claim that something directly causes another thing, you are expected to prove that there is some moment of time and space where the cause and effect come into direct contact with each other, or at least to demonstrate that this is a highly reasonable assumption to make. Scientists extend this concept with a sort of shorthand: the two objects may not really bang into each other, but then they have to be linked by steps such as a transfer of energy that do follow this rule. So to make a scientific claim that a child inherits traits from its parents, you have to find some direct mechanism linking them, such as the DNA in their cells. It is directly passed to the oocyte from DNA from the reproductive cells of the parents, and gets copied into each cell, and then it gets used in the transcription of RNAs and translation into proteins through a lot of single, physical interactions. You’ll never directly see all of those things happening, but the models you use predict they are there.

The second principle applies this type of causality to entities as complex as organisms or entire ecospheres. It shows what happens when a lot of local interactions create systems that are much more complex. At that point the principle declares that the state of a system arises from its previous state through a rule-governed process. From that it follows that future states of the system will arise from the present one, following the same rules. We’re far from knowing all those rules, but scientists assume they are there, and a lot of their work is aimed at creating models that describe them.

Both of these concepts are closely tied to a style of argumentation that integrates Occam’s razor; I’ll talk about that elsewhere.

How are these fundamental principles linked to evolution? Well, you start by observing what is going on in a biological system right now and creating models that project the state into the past and future. You test those models with experiments, and then start extending them farther and farther into the past and future. You make predictions about what will happen if the model is correct in the future, and look for evidence of its activity in the past. If something in an experiment violates those predictions, you have to revise the model. This process of observation, modeling, and challenging models is the source of the Big Bang theory in astrophysics; it’s the basis of our geological understanding of the Earth’s crust, and when Darwin applied it to life he got evolution.

Other belief systems such as religious accounts don’t start from an assumption that models are works in progress that will inevitably be revised; nor do they require that their versions of things constantly be revised to conform to evidence. It leaves people free to believe whatever they like, to maintain idiosyncractic positions in the face of mounting evidence to the contrary. It leads to inconsistencies about the way they think about causes and effects in their daily lives versus how they extend their opinions to the universe. This is pretty egocentric; it leaves no place for self-doubt and encourages no respect for the potential validity of other belief systems. This very easily slides into a type of intellectual authoritarianism which is absolutely counter to the fundamentally democratic nature of science.

You can see these two principles at work in the way we distinguish “scientific models” from every other kind. Anything that violates the principle of local interactions would be considered non-scientific. That’s the case for extrasensory perception – until someone demonstrates that some energy passes from one person’s mind into another’s, you can’t make a scientific claim for its existence, so you have to look closely into whatever model of causality led you to claim it might exist. And the second principle implies that there are no discontinuities – you can’t create something from nothing. Miracles and the fundamentalist account of creation violate both principles.

If you can’t agree on these two things, it makes very little sense to discuss details of evolution that derive from them, because the differences in the very basic assumptions held by people can’t be resolved – you’ve got to agree on things like standards of evidence and causality. If you don’t do that you can’t even agree on the meaning of words. That’s what makes these fundamental principles ghosts in “debates” on evolution, and they are the things you need to clarify before getting involved in one. And, of course, you have to insist that the participants act in a way that is intellectually fair and honest, with integrity.

There are a lot of other debates in science – such as controversies over animal experimentation – in which this doesn’t happen. Reputable organizations make inflammatory remarks and hold untenable positions on points of fact, and refuse to back down when you refute their points. Then you get barroom brawls rather than civil discussions about important topics.

You came up with this concept of “ghosts” while working on texts by students and other scientists. Why are they a particular problem for students?

An active researcher is usually so deeply engaged with his models that they have become a fully natural, shorthand style of thought. It’s like the grammar of a native language, which becomes internalised without a real understanding of its structure. In science this grammar has to do a lot with models. Most projects in research take place in a fairly exact dialog with specific models you are either trying to elaborate on by adding details, or extend to new systems, or refute through new evidence. This makes models very dynamic, and there’s no single reference on the Internet or wherever where you can go and find them. In biology virtually every topic gets reviewed every year or two, which is an expert’s attempt to summarize the most recent findings in a field to keep people in a field more or less on the same page. That’s the group that a lot of papers and talks are addressed to, at least most scientists think that way – and they assume the readers will have more or less the same concepts, models and frameworks in mind. Anything that is widely shared, people often fail to say – they think they don’t need to. And it’s impossible to lay out all the assumptions and frameworks that underlie a paper within it – you can’t define every single term, for example. So these become ghosts that aren’t explicitly mentioned but lie behind the meaning of every paper. The two really huge basic principles I mentioned above are rarely, rarely described in papers.

And even the details of the models more directly addressed by a piece of work – the physical structure of the components of signaling pathways, or all the events within a developmental process – aren’t mentioned very often. Those models are embedded in higher-level models, and the relationships in this hierarchy are not only hard to see – there’s no single way of explaining them. Scientists sometimes work these things out fairly intuitively as they extend the meaning of a specific set of results to other situations and higher levels of organization.

Now imagine a science student who is absorbing tons of information from papers like these. As he reads he’s grappling with understanding a lot of new material, but he’s also actively building a cognitive structure in his head – I call it the “inner laboratory, or cognitive laboratory.” It consists of a huge architecture in which concepts are linked together in a certain structure. The degree to which he understands a new piece of science depends on how that structure is put together, and where he plugs in new information. If the text he’s reading doesn’t explicitly tell him how to do this, there will be a lot of misinterpretations.

How can his professor or the head of his lab tell whether a scientist under his supervision is assembling this architecture in a reasonable way? You catch glimpses of part of it in the way someone designs an experiment, but I think the only method that gives you a very thorough view of it is to have the young scientist write. That process forces him to make the way he links ideas explicit and put them down in a way you can analyse each step. In writing – or other forms of representation, such as drawing images or making concept maps – you articulate a train of thought that someone else can follow, providing a means of interrogating each step. Most texts are pretty revealing about that architecture; if you read them closely you can see gaps, wrong turns, logical errors, and all kinds of links between ideas that a reader can examine very carefully.

The problem is that in most education systems in continental Europe, in which most of the scientists I deal with were educated, writing is not part of the curriculum. Whatever training they have is done in all sorts of ways, and the teaching is usually not content-based. Instructors use all kinds of exercises on general topics, but that learning doesn’t transfer well to real practice. Why not? Because when you write about a general theme, your knowledge is usually arranged very similarly to that of the teacher’s and any general audience. In your specialized field, on the other hand, your knowledge is likely to be very differently arranged, and that’s where the ghosts start to wreak real havoc on communication.

So ghosts aren’t just things that scientists leave out of texts – they’re also phenomena that arise from the reader or audience…?

Absolutely – they arise from differences in the way a speaker and listener or a writer and reader have their knowledge organized. That can happen in any kind of communication, but in science it’s actually possible to pin ghosts down fairly precisely. In political discussions or other types of debates there aren’t really formal rules about the types of arguments that are allowed… But if you know how meaning in science is established, you can point to a specific connection in a text or image and say, “To understand what the scientist means, you have to know this or this other thing.” Again, since neither of you can directly see what’s in the other’s head, a reader may not guess that some of the meaning comes from very high levels of assumptions, or a way of organizing information that you’re not being told. And some have been digested so thoroughly by scientists that they’re no longer really aware that they are there.

Some of the most interesting ghosts appear when you try use someone’s description of a structure or process to draw a scheme or diagram. I recently had to draw an image of how a few molecules bind to DNA because we needed an illustration for a paper. I thought I had it clear in my mind, but I ended up drawing it five times – each version incorporating some new piece of information the scientist told me – before I got it the way she wanted it. You learn an incredible amount that way.

A scientific text is often based on an image of a component or process that a scientist has in his mind. He’s trying to get a point across, and to understand what he means you have to see it the way he sees it – but if he leaves anything out, it’s easy to completely miss the logic. It’s like trying to follow someone’s directions… That works best if the person who’s giving the instructions can “see the route” the way it will appear to you, maybe driving it one time to look for the least ambiguous landmarks, or taking public transportation and watching exactly what signs are the most visible. And thinking it through with the idea, “Now where could this go wrong?”

Another thing you refer to is concept maps – you include several examples in the article. How do they fit in?

Concept mapping is a system invented by a great educator named Joe Novak; it gives you a visual method to describe very complex architectures of information. It’s extremely useful in communication, teaching, and analyzing communication problems. One reason it’s so important is that our minds deal with incredibly complex concepts that are linked together in many ways. Think of trying to play a game of chess without a board – that’s incredibly difficult, but a chess set is a fairly simple system compared to most of those that science deals with. There’s really no way to keep whole systems in your head at the same time. Making a map gives you a chance to see the whole and manipulate it in ways that would be impossible just by thinking about it.

But the real genius of this system appears in communication and its most precise form – education – where a teacher ought to understand what he is really trying to communicate, and how it’s likely to be understood by the students or audience. In most cases you’re hoping to do more than just “transmit” a list of single facts; you’re trying to get across a coherent little network of related ideas, linked in specific ways. If you do that successfully, the audience will leave with a pattern they can reproduce later. It might be a story, a sequence of events, or a metaphor – the main thing is, they have seen how the pieces are related to each other.

A great way to do this is to make a map of the story you’re trying to tell, and then make your best guess about how this information is arranged in the heads of your target audience. What can you realistically expect them to know, and what information and links are likely to be new? If you see the pattern you’re trying to communicate very clearly, and make a reasonable guess about how some type of knowledge you can relate it to is arranged in your audience’s head, you know what you have to change to get them to see things the way you’re hoping. In schools they’re teaching kids to make concept maps early on. Then before a lesson about something like the solar system, the teacher has the kids draw a map of what they think about the sun, moon, planets, and so on. After the lesson the kids make a new map – comparing the two tells you what they’ve really learned.

In your article you point out ghosts that come from schemes like sequences of events or tables…

A lot of scientific models consist of sequences of interactions between the components of a system. Those start somewhere and involve steps arranged in a particular order, and it’s important for the reader to have a view of the steps and that order in his mind. You’d be surprised how often scientists describe these processes in some bizarre order that doesn’t go from A to K, but starts at G, goes to H and I, then goes back to G and works backward to F, E, and D… Again, if you are already familiar with the sequence or pathway this is no problem. But if you don’t, you’re probably expecting the reader to try to assemble the process in some reasonable order. That may be possible through a careful reading of the text, but it takes far more “processing time” than a reader would need if the whole sequence were simply laid out in order in the first place.

Tables are interesting because a lot of experiments are designed with a structure that’s pretty much inherently that of a table. Say you have two experimental systems plus a control, and you apply two procedures to all of them. To make a claim about the results, you have to march through all these cases – basically a table that’s 3×2 or 2×3. Here again, you’d be surprised how many scientists’ descriptions skip over some of the cells of the table, mostly because the results aren’t very informative. Or they tell you, “Procedure A caused a 5-fold increase over Procedure B,” without telling you what happened in the control.

Both of these effects are due to a scientist’s failure to recognize the structure of the information he has in his head and is trying to present… Then he fails to present that structure in the text in a way that’s easy for the reader to rebuild in his own head.

You’ve said that ghosts are one component of a larger model you’re working on that reformulates the relationship between science and communication… What else is there?

A lot of the other points can be captured through an exploration of what I call this “inner” or “cognitive” laboratory of science. The really good scientists I know have a very clear understanding of their own thinking. They know the assumptions that have gone into the models they are using, and are aware of the limitations, where there are gaps and so on. That type of clarity usually translates into good communication, no matter what the audience.

One thing I found during this project that was very surprising was the extent to which writing and communication for all kinds of audiences was connected, and how addressing very diverse audiences could clarify thinking in a way that improved a scientist’s research. When you find a scientist struggling with clarity in a text, it usually means one of two things. Either a topic is not clear in his head at that moment, or it’s not clear in anybody’s head at this moment in science… That second case is very interesting because it means you can find interesting questions just through a very careful reading of a text, realizing that it’s asking you to build a certain structure of ideas. If you have difficulty, that means something. One of the basic strategies I used in working these things out was that problems are meaningful – they’re trying to tell you something about how good science communication works, or how scientific thinking works… usually both.

Speaking to a general public with really no specialized knowledge of a field can be a truly profound exercise for a scientist. It makes him interrogate his own knowledge in alternative ways. He has to come to a much more basic understanding of the patterns in his inner laboratory and apply different metaphors, trying to map that knowledge onto someone else’s patterns. Well, the cognitive laboratory is already metaphorical, based on concepts rather than real objects, and applying new patterns or metaphors to what’s in there is extremely interesting. It can suggest questions you’ve never thought of before. This means that tools that have been developed by linguists and communicators can be used as tools to crack open scientific models.

I’ve actually done this – used those tools to expose an assumption about evolution that everyone was making but wasn’t usually aware of. The assumption had never been tested, so my friend Miguel Andrade decided to take it on as a project, and put a postdoc on it. The results were really interesting, showing that there were a lot of cases where the assumption didn’t hold – and we got a published, peer-reviewed paper out of it. That was three years ago, and in the meantime I’ve been involved in a number of similar projects that have had a similar outcome. A communicator who pursues questions about meaning and language has a different set of tools to understand how ideas are linked in scientific models. You’re freer to apply slightly different metaphors and patterns to ideas; you may be more rigorous in perceiving assumptions; metaphors and other tropes help you see cases in which people are reasoning by analogy rather than strictly adhering to the system at hand.

So these ideas aren’t just a way to help people plan and communication better – although they certainly help in those tasks. In fact they are much more fundamental in scientific thinking. Understanding these relationships between communication and science is a pathway to doing better research, through a better understanding of its cognitive side. I’ve noticed recently, for example, a lot of cases where the way people are thinking of complicated processes is drifting away from the language they use to describe them. The language is conservative and it may be hard to adjust. But that will be essential as the models these fields are using move forward and become so complex that our minds – and our language – may not be truly able to capture them.