Skunkworks1 refers to teams that are given the freedom to operate without strict rules or to groups of developers with special authority.
There are several vivid examples when, within a large corporation, a traditional scientific field, or an established market, teams emerge that act as a competing force, challenging the status quo and driving innovation from within.
Besides the footnoted Lockheed Martin, classic examples include Apple's Macintosh Division, Sony's PlayStation, Amazon's Prime Video, GE Healthcare's 'Imagination Lab,' IBM's Watson Health, and X by Google (now Alphabet Inc.).
We won't delve into these in detail here. For those interested in learning what common elements lie under the hood of similar processes across various fields of life and even exploring the 'innovation equation,' the book by physicist and entrepreneur Safi Bahcall, Loonshots: How to Nurture the Crazy Ideas that Win Wars, Cure Diseases, and Transform Industries, is highly recommended.
If we had to name just one field where such approaches work on an industrial scale, it would likely be cybersecurity, with its in-house and independent startups, ethical (and not-so-ethical) hackers, red and blue teams, and, of course, Bug Bounty programs.
We discussed this and a couple of related topics with the R&D team at Guardora: ML engineers, cryptographers, and mathematicians.
1. Do you feel like you're part of such a team, granted the freedom to play without rules? In your opinion, how much of this is romanticized, and how much is grounded in reality?
Both yes and no. On one hand, our team is working on a problem that currently has no standard solution that satisfies all stakeholders. This undoubtedly involves innovation and thinking outside the box. On the other hand, we are constrained by the acceptable level of computational resources our solution can utilize, as required by our clients. Additionally, we must adhere to corporate and government regulations, which any of our proposals must meet if we want them to be implemented in practice.
Looking at it more broadly, in our case, it's not just about the right to play without rules, but rather the necessity, or even the obligation, to do so due to the lack of established rules in the area we've chosen as our field of activity. And while the romantic aura undoubtedly surrounds such work, it faces daily collisions with reality, because the product we create must successfully fit into existing practices for developing and applying ML technologies.
Skunkworks is definitely about us! In the IT world, there’s a lot of fluff surrounding the product creation process. At this stage, we have only one important criterion—the usefulness of our solution. The reality here is that breakthrough ideas are less likely to emerge within large IT players and more often come from eccentric startups, which are later acquired by the bigger players. So, at times, pure creativity requires the removal of constraints.
Indeed, being restricted by rules doesn't feel like us at all 😊 We have a fantastic team, and the topic we're working on is highly relevant. If there's anything to improve, it would be to wish for even more drive and enthusiasm.
2. Is breakthrough R&D different in established fields, where a critical mass of accepted approaches has accumulated, compared to R&D on the cutting edge of science? In the first case, you need to find a breach in clearly visible castle walls, while in the second, you're navigating through the fog, constantly feeling your way along the bottom.
Yes, more or less. This is partly due to the forced unconventional approaches to problem-solving mentioned earlier. Often, even experienced ML developers find it quite challenging to grasp the essence of our approach during discussions, as it may seem impossible within the framework of established R&D in ML.
However, there is definitely a similarity as well—both cases require a substantial foundation of accumulated knowledge and experience.
What distinguishes the second from the first is the disregard for authority. One shouldn’t abandon an idea just because someone authoritative says it won't work. At the beginning, it might not even resemble science. Often, new directions are considered pseudoscientific branches. Accepting that a discovered solution might turn out to be a 'nothing burger' (the fear of failure) shouldn’t hinder the free flow of thought.
It all depends on the person. If someone has identified a problem and wants to solve it, they will find a way to do so, regardless of how established the field is. Moreover, 'navigating through the fog' can even be more interesting! Overall, to engage in cutting-edge R&D, one needs to be bolder and more curious. It's like going mushroom hunting—you can stick to well-trodden paths and collect mushrooms missed by others, or you can venture into uncharted territory. The second option carries more risk, but in both cases, you might end up with nothing or a basket full of mushrooms.
3. How often do you use metaphors? Do analogies from nature, pop culture, or other sciences provide you with useful insights for your work?
It seems that metaphors can be helpful when explaining our approaches and ideas to non-specialists in mathematics and cryptography. For specialists, more precise and formal formulations are more familiar and understandable. Therefore, within the development team, we do not seriously use metaphors and analogies, but we do frequently use them when communicating with non-technical specialists and clients.
We use these expressive tools with humor within the team when discussing unexpected results or brainstorming ideas. This is also useful.
It is hard to recall when examples from nature have been directly useful for our work. However, breakthroughs in research often happen in this way. For example, neural networks were not invented out of thin air but by analogy with the functioning of neurons in the brain, that is, borrowing knowledge from biology.
4. Negative numbers have no direct analogues in real nature. How do you deal with such abstract concepts that, at first glance, seem 'unnatural'?
Those who have tied their professional activities to exact sciences perceive negative numbers and similar entities without natural analogues as quite natural, without even thinking about it. Professional deformation, perhaps. 😊
The brain took a long time to get accustomed to mathematical thinking during the learning phase. Now it is easier to think, perceive, and articulate within specific structures. Problems sometimes arise when explaining one's thoughts to outsiders.
Mathematics, in general, consists of a set of abstract concepts. However, their 'unnaturalness' allows us to build various conclusions that cannot be made based on real-world objects. Periodically returning from abstractions to real objects to verify the conclusions drawn in the abstract world helps to stay connected to reality.
Accept negative numbers as they are. 😊
5. Among representatives of creative professions (and R&D is a creative field), there is a well-known saying: 'Inspiration is for amateurs. Professionals just take action.' What is your balance between discipline and talent? Are frameworks, processes, and metrics necessary for R&D?
Balance is necessary in everything, and this question is no exception. No matter how talented an employee is, they must have the discipline to bring their ideas to at least a working prototype. Similarly, no matter how disciplined and hardworking an employee may be, they must be able to generate ideas worth testing, rather than just waiting for assignments from others. The introduction of frameworks, processes, and metrics in R&D probably becomes relevant when the team grows so large that the tech lead doesn't have time to personally speak with each team member and discuss their tasks within a week. For smaller teams, such implementation is less relevant and only takes away productive time from employees.
Behind every creative endeavor is a certain foundation of preparation. The broader and higher quality it is, the more successful the creative process will be. Well-established processes allow for quick resolution of routine tasks and sharing the results with the team. Solving routine tasks and testing standard hypotheses is an integral part of R&D.
Some people perceive external control tools negatively, as something secondary and distracting. Their inner critic already handles the job quite well and rarely ends the day with a positive evaluation of their own results. The danger of metrics is that team members may start to hold back their results and ideas for the sake of personal achievement, which could negatively impact the product. It's understandable that this is common in the industry, but it seems that metrics are tools suited for organizations with more complex hierarchies and management structures.
For some, inspiration is key. Can you force someone to come up with something new? Probably not, as it must come from within. A person must be ready for a breakthrough.
6. Do you prefer talented team members or disciplined ones?
Perhaps the more important criterion is responsibility. If an employee approaches their task responsibly, it doesn't matter whether they solve it during work hours or at night, whether through inspiration or by sheer willpower and perseverance. What matters is the confidence that they will either solve the problem or, at the very least, try every possible way to solve it.
Honesty and openness to dialogue are preferable. Talent is a subjective concept, and discipline alone cannot guarantee results. But openness in discussing the team's challenges, honesty about the outcomes, and a willingness to share successes and failures with the team help everyone find the right path more quickly.
In general, this is a very complex question. Few people like either talented slackers or diligent slow-thinkers. Both are harmful, but if you had to choose, the preference would probably be for talent, as long as they are used as idea generators without the expectation of them handling the execution.
So, talent probably wins out. Although it's hard to imagine how you could measure the degree of talent. That's why we love all team members equally. 🙂
7. How do you recognize a brilliant specialist or a brilliant solution?
It seems that such conclusions should never be made hastily, relying only on a person's self-assessment or reviews about them, nor based on superficial information about a solution. Only through deep and detailed analysis of the solution or evaluation of the specialist's completed work can one form a certain opinion.
Making premature judgments is always risky. It takes time to understand both the person and the solution.
Some people have a natural ability to sense it. It’s hard to explain, but a good specialist (whether they are a researcher, a doctor, or even a consultant in a store) can always be felt — you trust them right away, especially when they answer your questions even before you've asked them.
8. On one hand, it seems that those in exact sciences are less susceptible to cognitive biases since they often work with experiments, evidence, and quantitative data. On the other hand, it is known that the roots of such biases lie in the emotional sphere, which acts instantaneously. How does this work for you?
Yes, in our field, it’s quite straightforward: as soon as there is confidence that an idea works, it needs to be experimentally tested, and everything falls into place. Occasionally, there are difficult cases where repeated experiments indicate that an idea does not work, but the author repeatedly searches for errors in the experimental procedure, remaining convinced of their correctness. Fortunately, we don’t have such cases in our team. 😊
In our case, the trap lies in complexity. When faced with sufficiently complex new information that you cannot immediately grasp and understand, there is an internal simplification that is extremely dangerous.
Certainly, conclusions from experiments are also based on subjective impressions. But it’s not always possible to verify whether these impressions are truthful (based on experience) or not (possibly cognitive biases).
Everyone has days when emotions decide everything. 😊 However, it is important to try to make all significant decisions in a state of emotional balance, relying on your inner feelings (intuition) while considering all available facts.
9. In popular science, you sometimes hear that the laws of physics 'prohibit' or 'allow' certain things. In this sense, mathematics seems less conservative. When you formulate hypotheses for testing, do you often estimate what is doomed to fail according to accepted norms and what has a chance? Is this based on calculations or intuition? Does this mean that you know all the laws of your science?
When setting up experiments, we certainly rank the hypotheses to be tested according to the expected payoff. When there are many hypotheses to test or limited time for experiments, ranking is often done intuitively. However, intuition is heavily influenced by accumulated experience, so these are not entirely unconscious decisions. It is impossible to know all the laws of mathematics; situations frequently arise where experimental verification of a hypothesis goes beyond expectations or even presents surprises.
Any hypotheses must be experimentally tested, and any anomalies discovered can be explained within existing norms; the issue is more about the depth of problem analysis.
There is an opinion that mathematics is not quite a science but rather the language spoken by exact sciences. Our practice in ML is a very specific applied branch of mathematics, which is heavily influenced and altered by computational capabilities. Therefore, often, internal assumptions of 'this will work' must be tested experimentally rather than being proved theoretically. ML remains more of an art than a craft, with the trajectory of knowledge being: hypothesis -> practical confirmation -> strict justification -> publication.
Indeed, mathematics is more flexible 😊 If there is a statement, such as 2+2=4, you can circumvent it by creating conditions where it does not hold. Continuing the example: 2+2=11 in a ternary system or 2+2=10 in a quaternary system. One can first rely on intuition and then seek confirmation of hypotheses. We don’t know all the laws for sure (and believe that it is essentially impossible).
10. The aforementioned Bounty practice in cybersecurity involves subjecting your developments, essentially your 'creation' to various attacks, with successful attackers receiving significant rewards. Does this excite you with a sense of competition, or would you prefer to avoid the associated anxieties?
This is an extremely useful and more than effective approach in analyzing the resilience of software protection against malicious actors. Developers of security often have no time to keep up with all the innovations in attackers' arsenals published in specialized communities and literature, and it is better to identify vulnerabilities at an early stage rather than dealing with the consequences later. However, of course, a separate emotional reward for the security developer is the fact of a fruitless bounty, indicating that everything was done correctly. 😊
This is a normal practice in the field of information security. Here, the product developer is always more vulnerable than the attacker. If an attack succeeds, the developer is at fault, while the attacker is praised. Otherwise, no conclusions can be drawn except that the specific attacker failed, which does not give the developer a sense of peace. The only concerns are that attackers operate within the threat model we assume for the product.
In general, playing in an unregulated field is inherently anxiety-inducing. And if we cannot test our developments ourselves through a regulated set of rules, then passing them to all willing testers is the only way to confirm or refute their reliability and security.
After all, any creator wants their ''creation' to be perfect (or as close to perfection as possible). Therefore, constructive criticism that allows improving the product should be seen positively as a push towards further development and enhancement of the product. Although in practice, it is, of course, impossible to avoid anxieties and frustrations.
11. It is said that mathematicians working with abstractions can sense the irrational. Do you think there is something that remains beyond explanation even in the language of mathematics, let alone in plain human language?
We cannot speak for everyone, so it's better to refrain from unsupported personal conjectures.
In this case, it depends on the explainer. Individual experience suggests that the more a specialist understands their field, the easier it is for them to explain something not only to their colleagues but also to people who are far removed from the subject area.
Mathematicians are people with their own feelings, beliefs, and religious convictions. And since even in this they have not been able to reach a common denominator, it is quite legitimate to assert that mathematics is incapable of describing the world of feelings. Those who work in mathematics are simply more critical of their conclusions, leaving room for the possibility of errors in the axioms from which they are derived.
There may indeed be something inaccessible… Or perhaps we simply have not fully explored the language of mathematics yet. After all, any language evolves with the introduction of new objects and phenomena in our lives that need to be described in some way.
12. Is there any irrational factor influencing the outcome in your field?
There is such a factor, and that factor is faith in successfully solving the task at hand. Without this belief, neither discipline nor talent can help achieve the result. The belief that we are on the right path provides motivation, makes us search for ideas in places where everything seems already explored and prevents us from being discouraged when another idea fails to deliver results.
13. Does the story of Srinivasa Ramanujan, who often relied on intuition in his mathematical discoveries, inspire you? How important is intuition in your approach?
Yes, his story is remarkable, but his discoveries were not based on intuition but on the unique structure of neural connections in his brain. He was a one-of-a-kind person; others simply don’t have the same capacity. What is difficult for a person to logically explain to themselves, they often attribute to chance, intuition, fate, and so on. In mathematics, intuition is often ascribed to what the author has been working on for a long time, processing information in their subconscious. It is said, according to legends, that many great pioneers dreamed of their major results in their sleep.
The case of Srinivasa Ramanujan is both awe-inspiring and intimidating. In our experience, it is preferable to link intuitive assumptions to the successful putting together of the puzzle of knowledge in the subconscious depths of the mind.
We admire people who deeply sense patterns. Whether intuition is important or not depends more on the individual than on the approach. One should use what they are accustomed to relying on in their discoveries.
14. Quantum computers can break any passwords—Is this true or an exaggeration? What do you see as the potential of quantum computing in the field of security?
Certainly, this is an exaggeration. Many encryption algorithms are already standardized and proven to be resistant to attacks performed by quantum computers. However, it is important to note that current widely used information protection algorithms are not resistant to quantum computers, and transitioning to quantum-resistant algorithms requires time, resources, and willingness. Therefore, if a quantum computer with 256 qubits appears tomorrow, its owner could gain access to a vast amount of protected data, from banking transactions to personal messages in messaging apps.
Quantum computers are not universal and can only provide a computational advantage in a specific range of tasks. A broad spectrum of information protection algorithms resistant to attacks, which quantum computers might offer a computational advantage over, has already been developed. Nevertheless, the advancement of quantum computing could lead to breakthroughs in many areas, and Machine Learning is no exception.
Quantum computers pose a threat to a quite limited set of information protection methods. Even in these cases, the breach would not occur instantly but might result in at most a quadratic reduction in brute-force search time. Cryptographers have long been preparing to counter the "quantum apocalypse" and can already offer techniques resistant to quantum computing. More interesting are the applications of quantum computers in AI. How will this impact the industry and what will it lead to?...
1 The term "skunkworks" originated from Lockheed Martin's advanced development program known as Skunk Works. It was created during World War II to rapidly develop innovative aircraft outside the usual bureaucratic processes of the company. Skunk Works was responsible for some of the most groundbreaking aircraft designs, including the U-2 spy plane, SR-71 Blackbird, and F-117 Nighthawk, the first operational stealth aircraft. The team was given autonomy, resources, and a mandate to innovate without restrictions from the main corporate structure. This allowed them to act as internal competitors, pushing boundaries and thinking outside the box.
