Join gaming leaders online at GamesBeat Summit Next this upcoming November 9-10. Learn more about what comes next.
Since Dell went private, it doesn’t have pressure from Wall Street to show a profit at every instant. The computing giant can invest in its own long-term research projects, and it has been doing so for some time at research offices in Silicon Valley and at its headquarters in Round Rock, Texas.
Jai Menon, chief research officer at Dell, is in charge of directing Dell Research into high-priority areas. He has focused on building user analytics and interfaces that can divine more about a user’s mood and exactly what their intent is when performing computing tasks.
His teams are also delving into security that works better because it is context aware. It can, for instance, use face recognition to figure out if you’re in front of your computer and lock it down once you’ve left the vicinity. If someone else comes up to your computer while you’ve left, it will tell by the way that person types that it’s not you.
To get computers to do this kind of intelligent work, they have to get much better at sensing the environment around them — and even sensing biological information about you — and becoming aware of the context of your work.
Menon believes that computers will become more specialized and become predictive and prescriptive over time. They will sense things in real time and enable new types of mobility. We recently met with Menon at the sprawling Dell Research office in Santa Clara, California, where a lot of this work is being done. Here’s an edited transcript of our conversation.
VentureBeat: Tell us about what your charter is at Dell Research.
Jai Menon: We have a charter to work on organic and long-range and disruptive kinds of research. We also produce what we call the Dell Technology Outlook. That’s our point of view on where the world is going over the next five years. What are some of the emerging trends? It’s a Dell point of view. We take input from a lot of sources inside and outside Dell, but at the end of the day it’s our Dell point of view on what’s happening.
We use that to drive our organic research. If we think X is going to happen, we want to be working on X in research. We also use that to influence Dell’s strategy. We want not just Dell Research, but Dell overall to be moving in the direction of what our DTO is talking about.
We work with universities as our third charter, drive university alliances around research, and then the fourth charter is to be the sponsor of innovation across all of Dell. Not just within Dell Research. There’s lots of innovation happening in other parts of Dell. We want to be encouraging and sponsoring that.
All these things are bucketed into four areas of work. My group has four sub-groups or four themes for all of our research. One is modern and next-generation infrastructure in the data center. Another is big data and data analytics and data insights. A third is around security. The fourth is around mobility and the Internet of Things. The next-generation user interface falls in that fourth bucket. All our thinking and projects and the DTO, all these things are focused around those four areas. We have key researchers in each.
One of the ways I measure the value and success of Dell Research is how we influence what Dell does. One area where we’ve had success since the time we last talked, about a year ago — I think we talked about the fact that we were looking at a specialized cloud focused for the telco space. That work made a lot of progress. Dell created a new business unit to go after that opportunity with the telcos. We announced the first set of products and offerings from Dell around that in October. That’s a proof point, to me, that the work we’re doing and what we talk about in the DTO — we’re able to influence Dell to create new product offerings. Meanwhile we’re continuing to drive that high-velocity cloud work forward from a research perspective.
VB: Can you tell us more about the kind of work that business unit is doing?
Menon: The telcos today use specialized hardware, custom hardware to run their networks. All the mobile traffic in San Jose goes through some evolved packet core gateway-type hardware devices. What we were able to show is that instead of having these special hardware devices, you can do as well or better from a performance point of view, and make it much cheaper, if you use standard servers from Dell and the function is implemented in software on the server.
You don’t need the custom hardware. You can still push through the same packets per second that the special-purpose boxes were pushing. You can handle all the communication in a million-person city like San Jose or San Francisco. They can all be talking to each other or video-chatting with just a quarter rack of Dell servers and networking here. You don’t need a quarter-million-dollar box that they’d otherwise have to pay for.
What we announced as products are these integrated offerings that have Dell servers with the appropriate levels of software to handle these kinds of workloads. The packet core handling mobile traffic, all the things the telcos need to do. We’re offering a solution in that space. They can just get that from us now instead of buying specialized hardware.
VB: What’s the future outlook like?
Menon: We’re going to be releasing, fairly soon, our overall Dell Technology Outlook, talking about the trends we see in each of those four areas I mentioned. I’ll focus on one of those four, given your particular interest. Our point of view is that user interfaces will evolve to not just know things about you from a physical location perspective — things like GPS and so on — but will get to know more about the user’s emotional state and other things like that. That will help us do a better job on behalf of the user.
Right now our system works at 30 frames per second, reading your mood through your facial expressions 30 times a second. That’s purely for demo purposes. The other thing it’s doing, do you see these dots here? One of them is your eyes and one is where your nose is pointing. We have eye tracking that’s going on here as well.
VB: So this works on audiences?
Menon: We have another version of this. This is a single-user version, but we also have a crowd version, which we used when Dell sponsored a TEDx conference in Amsterdam last November. We used it to get a sense of how the speakers were doing. We trained it on the front two rows of the audience and cataloged the reactions of about 20 people. It was good feedback for the speakers. With the facial recognition, we can also do demographics — male or female, young or old — and so we could give them feedback like, You’re doing well with women in the audience but not so well with the men. They liked it so much that they want us to do more of this for all the TED conferences.
The idea we see is, if you’re playing a game and I can tell that you’re bored, I can ratchet up the challenge level. Or in a work setting, if I can tell you’re concentrating very hard, I can change the settings on your phone to go straight to voicemail and not disturb you. If you’re a driver, or if you’re on a manufacturing floor, if I can tell that you’re not attentive right now, then it may be time to take you off the assembly line or have you stop driving.
We think there are many practical applications of sensing emotions, sensing attentiveness, and sensing demographics. If I can tell that you’re a child, I can protect you from going to certain places on the Internet. If I sense that you’re an older person, I can immediately increase the font size of what you see on the screen. This is why we’re doing this type of research.
We also have a version that’s based on voice. When you speak to it, based on the stress levels in your voice we can tell a few things. Our goal is to look at all the different ways in which we can read you and get as good a performance as we can from the different possibilities. We also have some of these new neural devices that read your brain waves. We’re doing some experiments with those to see how well it translates into being able to tell your mood. There are multiple ways to get information about you, and we see multiple ways to leverage that information.
VB: Do you get any help from things like wearable devices — monitoring heart rate and so on?
Menon: Exactly. We want to take all of that and integrate it. Part of our value will be to figure out which are the best and what’s the best way to integrate all these things to get knowledge about you and figure out the way to use that knowledge to support you. That’s the general thrust of this area of work.
VB: How far along would you say that work is? Has it been going on for many years already?
Menon: I’d say there’s a couple of years research gone into it so far. We’re working with partners. We’re just now hiring someone with a PhD in this area who’ll start with us in April.
VB: How does that work with folks like, say, Microsoft, who make the operating system, or Intel RealSense?
Menon: The quality of the camera is directly a function of how well you can do the spatial stuff. Having partners like Intel, with their really good cameras, we think that can enhance the quality of what we’re doing here. Obviously Microsoft and Intel are good partners with us. We have good connections into their labs and so on.
VB: Where would Dell be best at using some of this? In the enterprise in some way? Customer support?
Menon: We have the Alienware offerings. In a game setting, we could deploy it there for detecting changes in challenge level. The customer support issue is a good one that we’ve talked about. Partly because this is still in a research phase, we don’t want to get too far ahead of ourselves with respect to all the business possibilities. We want to get it to a point of, what are the areas it’s going to be really good at, what are the areas it might be good at, before we get more deeply into the specifics of how we might deploy it.
VB: Is this something you’re targeting for around 2018?
Menon: Yeah, 2018 sounds like the right time frame for this one.
With respect to things like BYOD, a lot of the focus has been on the security aspects of making sure that somebody who brings their own device doesn’t do the wrong thing and so on. But there’s also a usability angle. We feel the next stage will be about combining security and usability.
What we’re showing here is that in this room, right now, you have both mobile access and Wi-Fi access. It’s telling you how good these things are. Wi-Fi is not so great. Mobile is in pretty good shape. We actually measure not just the signal strength. The five bars you see are just signal strength. You could be at Starbucks and it says you have five bars, but you’re not getting good performance because 100 other people at that Starbucks are using the network too. You have to measure things like latency and how well the signal is actually performing.
What this lets you do now is, with my application, I can go in and modify my settings. When I’m running my application I can say, “The default for all my applications is seamless.” Whichever’s better, use the one that’s better. If Wi-Fi is better, use it. We can walk around and I’ll show it to you. It will automatically switch over. You don’t have to do anything. No manual intervention needed. It just switches you over from Wi-Fi.
We also have aggregation as an option. If you have an app that needs a lot of bandwidth, take everything you have, add it together, and give me maximum performance. If you have a data plan and you’re paying for that, you could set up your application to prefer Wi-Fi. Whenever you have Wi-Fi, use Wi-Fi. Don’t use the cell connection because I’m paying for it. Or if you’re in Europe, they charge you an arm and a leg for cellular, so prefer Wi-Fi.
If you go over here, you can see how much we’ve used Wi-Fi and how much we’ve used mobile. If I run a little speed test here, you can see that it’s using a little bit of both. It’s seamlessly shifting back and forth. Sometimes it’s using a little Wi-Fi, a little mobile. Here, I can turn on a radio app. As the radio’s running, you can see that it’s mostly using Wi-Fi. Going back over here, whatever’s good is what it’s going to use. We can walk around a little bit and see that it starts to change what it ends up using. The whole time, the radio is still playing. It doesn’t flicker.
VB: What specific kind of research is that? Just signaling?
Menon: You have to measure the signals and communicate with the server. The server has to decide when to switch and when not to switch. You have to look at both signals and the latency to do it in a seamless way. My server that’s controlling this is in San Francisco right now.
What we’re trying to do here is as follows. We have this global technology adoption index, a survey of a lot of customers. Fifty percent of those customers said that the reason they’re concerned about using mobile devices is because they’re worried about them getting stolen, the data breach that could result from someone else getting hold of your mobile device and accessing your stuff.
This is a project called continuous authentication. Unlike authenticating yourself once at the beginning, by typing a password or giving a fingerprint or whatever, the idea is that I’m constantly evaluating — If this is your machine, is this still Dean using it? We do it through things like swipe on the touchscreen. The way I swipe up and down and left and right is different from the way you do it. The pressure I apply is different from the pressure you apply. Over time you want to look at other signals as well.
VB: So it’s the problem of what happens if you go away from the computer and someone else gets on it.
Menon: Or if you leave your device here and someone picks it up. They can get lost or stolen so easily. It falls out of your pocket. This would work with a mobile phone or a tablet. We’re looking at a combination of things, but right now we’re focused on gesture, swipe, and pressure patterns. Over time we can integrate facial recognition. That would be a two-factor authentication. Another factor is that the words I use are different from what you use when we write emails or stories. In any case, part of the message is that we see the world moving from one-time authentication at the beginning of the session to this continuous authentication.
VB: The facial part of that, how much of that is just continuously looking to see that it’s the same face?
Menon: The combination, I think, would be foolproof. Facial recognition alone is not foolproof. This swiping system may not be foolproof. But we think the combination, that would be nearly foolproof. Here, you can swipe up and down a couple of times and pretty soon it will recognize that you’re not me.
Long term, a lot of people are working on getting rid of passwords entirely. People really don’t like passwords. A lot of people don’t feel that passwords are very foolproof, and they’re kind of annoying. Maybe this combination of facial plus gestures is a way toward looking at the option of a password-free way to authenticate yourself.
VB: Would you call this context awareness?
Menon: We have a number of directions in that respect. What we’re trying to determine is, what is normal behavior and what is not? We’re also looking at the context in which you do things. Today the way things work is, “Jai is an admin, and an admin gets access to all kinds of stuff.” I want to change it to, “Jai gets access to stuff, but not if he’s currently using a non-Windows device and happens to be in Russia at the moment.”
That’s context-aware in the sense of, I’m trying to understand what are normal and abnormal behaviors. It’s not normal for you to be using an Android device from Russia. Using the context of how you’re using the device, what device you’re using, what the threat level is today, where you’re coming from, whether you’re inside the firewall, all these things factor in. We also want to factor in business context. If I know that Dell and some company just signed a big contract yesterday, maybe there’s a reason a bunch of money is getting transferred from here to there. If I know the business context, I may allow something that would otherwise seem suspicious. That’s what we’re looking at with the context-aware security work.
VB: Do you have any partners in that area? I know there are some folks who use keyboard analysis to figure out some things like, are 100 people using one subscription, or just one? Some of this stuff is available out there already. Where do you think you’re pushing forward into original work?
Menon: The continuous authentication stuff, using swipes and gestures, we’re pushing forward without partners. We think there’s a lot of research still to be done there. We’ve submitted a proposal to the Department of Homeland Security. They’re very interested in this problem and they’re willing to spend a lot of money. We’ve passed a first round with them. We’re in the top 20 percent of proposals. But we have to work our way. They have what they call a broad agency action, like a call for, “All you smart guys out there, tell us how to solve this problem.”
The work on facial recognition, there are some partners for some of that stuff. But a lot of it — Tying it to mood is interesting, tying it to emotion. The voice-based work is something that we’re driving here. Integrating all that is where — People might have said, “Let me try this, let me try that,” but there isn’t a clear state of the art as far as what really works, what combination of things is really foolproof.
In the area of encryption, there’s very significant research yet to be done about homomorphic encryption. The basic idea here is that we know how to encrypt data when it’s stored on a disc. You can encrypt data when it traverses the wires. The interesting question is, what about when you need to perform some operations against the data? When you bring the data off the disc and now you want to perform actions on it, you bring it into memory, and now you have to decrypt the data. It has to be in the clear so you can compute on it — add and subtract and do whatever you need to do.
VB: What do you call this?
Menon: What homomorphic encryption lets you do is, it lets you compute on encrypted data. You don’t have to decrypt and it still works. Let’s say that I have a tax accountant doing my taxes for me. I have to send him my salary information. I can send it to him encrypted, but he can’t run something like TurboTax unless he decrypts it and sees my information. Similarly, once he’s done, there’s a number for how much I have to send to IRS. He can encrypt that and send it to me so nobody else sees it, but he himself got to see it while it was in the clear at the end of the operation.
With homomorphic encryption, I can encrypt my salary and send it to my tax guy. Then he can do all his computation on encrypted data and send the results back to me, but he doesn’t actually know how much I’m sending to the IRS, because the result is encrypted too. Only I have the key. I can use a tax guy and he never gets to see my income or my tax.
That’s homomorphic encryption. It’s an area we’re quite interested in. We’re exploring that. There’s still about three or four years of research to be done to make it really practical, but that’s within the five-year horizon of what we’re looking at in Dell Research. Smart people in universities are working on it. We’re trying to partner with people like that to continue to make progress in this area.
We talked a bit about predictive security as well. The analogy I like to use is waiting until your house gets burgled and then changing the lock. That’s a little late, but it’s a lot like what happens today. Once you have an attack, you change the settings on your firewall to protect yourself from the same attack in the future. The next step would be waiting until your neighbor’s house gets burgled and then fixing your locks. We can do that, because we have two million Dell SonicWall appliances out there. If we can see a hack happening in India or China, we can change the firewall settings on all our other SonicWall appliances.
You really want to go even further, to the point of being more predictive. You want to see if you need to make the lock better even as the attack is taking place, as opposed to waiting for the attack to take place.
VB: Now that you have research underway in these areas, does that give you more optimism about security in the future? Its current state seems to be kind of dismal.
Menon: That’s one of the reasons why we started thinking around things like data leakage protection. That’s really hard, making sure that data never leaks out. That’s why we’re starting to think in a completely different framework about this. What if it doesn’t matter if the data leaks?
VB: Do you feel like we’ve been losing a security war, and that we can eventually start winning?
Menon: I wouldn’t say we’re losing the war. I think we’ve not yet connected the dots in a way that we could. That’s part of what we’re trying to do. When you have information about what’s coming through the SonicWall appliance, information on your identity and access management software — We’re starting to do things now, for example, where the client has encryption software. Let’s say that someone goes to Dropbox and starts to move some data out there. If you go through our firewall, we can see that you didn’t encrypt the data when you moved stuff out. Then we can automatically load the encryption software onto your client and get that done before it goes out.
I think there’s a lot of optimism in terms of what is possible. Those are some of the pieces that need to happen in order to win this thing.
VentureBeatVentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact. Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:
- up-to-date information on the subjects of interest to you
- our newsletters
- gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
- networking features, and more