The trajectory of innovation in AI today reminds me of the early days of the web and the browser wars that shaped it. Back then, our ability to create and interact with the web was constrained by a single browser, limiting the fidelity of what we could produce. As competitors entered the scene, standards like HTML and CSS emerged, and the web expanded exponentially. Even with those standards, new innovations like JavaScript and Macromedia’s Flash unlocked possibilities we hadn’t imagined, breaking through the constraints of what we thought browsers could do.

AI is heading down a similar path.

The Parallel to Browsers

We’re currently in an era where innovation in AI models is being driven by a mix of proprietary players (like OpenAI and Anthropic) and open-source challengers (like Meta’s LLaMA and the newly released DeepSeek R1). Much like the browser wars, this competition is defining early standards and pushing the boundaries of what’s possible. Proprietary models bring alignment, safety, and polished performance, while open-source models democratize access and drive experimentation.

But even as we’re watching this battle for dominance unfold, I think the focus on which model “wins” is irrelevant in the long run. Just as browsers settled into a multi-platform world with innovation layered on top, I believe we’re heading for a multi-model AI ecosystem, where different models—both proprietary and open-source—work together in a modular, interoperable way.

A Multi-Model World

The future isn’t about one model dominating. It’s about how we combine and integrate different models to meet specific needs. Proprietary models will likely dominate in commercial, consumer-facing products where safety and alignment are non-negotiable. Open-source models, on the other hand, will thrive in research, niche applications, and cost-sensitive projects, pushing forward innovation in areas that proprietary players might overlook.

What’s important here is that this trend line isn’t just about scaling existing architectures or improving fidelity. It’s about creating new tools, systems, and even paradigms—like JavaScript and Flash did for browsers—that will fundamentally shift how we think about AI applications. I’m convinced we haven’t yet seen the step changes that will truly unlock AI’s potential. These will likely come from new players or hybrid innovations that combine dense, sparse, symbolic, and neural approaches into something we haven’t imagined yet.

Standards, Middleware, and the Rise of Hybrid AI

If we take lessons from the browser wars, we can see a few trends emerging:

1. Interoperability Will Drive Adoption: Just as the web couldn’t scale without standards like HTML5, the AI space will need standardized APIs, formats, and protocols to allow models to work together seamlessly. Companies don’t want to be locked into a single model—they’ll demand orchestration layers that let them mix and match tools based on need.

2. AI Middleware Will Be Key: As multi-model workflows become the norm, we’ll see the rise of platforms that act as a gateway for managing and combining models. Tools like LangChain or AutoML are just the beginning—these systems will evolve to abstract away the complexity of choosing and managing models.

3. Hybrid Architectures Will Take Over: Just as browsers now integrate multiple engines (like JavaScript and WebAssembly), AI models will likely become hybrid systems that combine the strengths of dense networks, sparse MoE architectures, and symbolic reasoning. These modular systems will be able to scale while still delivering task-specific performance.

4. Innovation Will Outpace Standards (For Now): Before interoperability catches up, we’ll live in a fragmented space where innovation moves faster than standardization. This means we’ll see exciting breakthroughs but also challenges in integrating them effectively.

Irrelevance of Individual Models

Looking at this trajectory, it’s clear that the individual models themselves are becoming less relevant. What will matter more is the ecosystem: how models are integrated, how they interoperate, and how they solve real-world problems together.

The browser wars ended with multiple players coexisting because the innovation became more important than the competition. The same will happen with AI. Whether it’s DeepSeek, Meta, OpenAI, Anthropic, or the next unknown player, the trend line points to a modular AI world where tools work together seamlessly—and innovation, not dominance, drives the future.

The big question isn’t “which model will win?” but rather, “how do we build the tools and systems to make them work together?”

I believe in hypertext. When Tim Berners-Lee proposed the concept and helped bring it to life, we all became a part of the world wide web. The concept is simple and beautiful. I can reference anything and link directly to something with meaning and additional context to my original work. For many years I have blogged off and on and one of my favorite features has been to leverage the “hypertext reference” feature. Now, thanks to Google, I have been asked to remove this link from a company working on behalf of another company that is being penalized by Google for my reference to a page on their site.
Now, first let me say that VirtualPBX has never paid me to include the link either directly or through another company. They’ve never asked me to include the link. They don’t even know who I am. I simply found their page explaining how call forwarding works, thought it was a nice little reference that provided more context to my blog post, and was happy to send a curious potential customer their way for their efforts in creating the page in the first place.
Unfortunately, “due [to] the Google Penguin update these links are negatively impacting both of our sites by creating an unnatural link.”
Here’s the email I received from a representative at RemoveThisLink.com:

Hello,
I have been contracted to help with the optimization of the site http://www.virtualpbx.com/. In light of recent Google updates the site’s link profile needs to be cleaned up, so we would like to request that you please remove the links to our site located here:
http://darinarcher.com/blog/2007/09/
http://darinarcher.com/blog/mobile_web/
http://darinarcher.com/blog/mobile_web/mobile_work_phone_that_switches_to_personal_phone.htm
Due do the Google Penguin update these links are negatively impacting both of our sites by creating an unnatural link. Please know that your site’s integrity and business practices are not in question, it is just that the existence of the link itself causes Google to establish a relationship between our sites that looks unnatural.
If you have any questions about the request above, please contact at Virtual PBX. He can be reached at .
Thank you for your consideration,

@removethislink.com

I think this is a real shame. I get that there is a significant amount of SEO spam out there, but the fact that I received this email, given my intent in my blog post… well, shame on Google. This sure feels evil and controlling and against the culture, purpose and value of the Internet.

I recently purchased a Motorola Droid X Android smartphone from Verizon Wireless after having been an iPhone owner for a little over two years. I made the switch for a few reasons, the two primary being a.) AT&T’s network sucks and b.) upgrading to iOS 4.0 on my iPhone 3G rendered it useless due to performance issues. Before I go into my reasons, I’ll openly admit that I’m an Apple addict and have been since I was in middle school. I love almost all things Apple and dreamed of the day the iPhone would be built. In the beginning I had many complaints, but over time either adjusted to its deficiencies or Apple resolved them with new capabilities. That said, I began my journey with an android phone with as much openness I could and to be honest was a little excited to make the switch and see what it was all about.

Android User Experience is Clumsy

One of the immediate challenges with the Android ecosystem of hardware vendors is that they are all making modifications to the user interface and in some cases almost entirely changing the experience. This is unfortunate as it means the evolution of improving the operating system and user experience will be disjointed and slowed as each group will do their own thing thus dividing the total potential resources that could be focused on making it better.

Keyboard Accuracy

The first major issue I had with my phone was typing. The keyboard and touchscreen that comes with the Droid X responded poorly to my input. I find the auto-correction to be much less accurate than what I became accustom to with the iPhone and found my ability to type simple emails and text messages tedious and slow. I have been really disappointed by this as it seems we had made large strides in the phone world with tiny thumb keyboards becoming increasingly accurate and predicting what we intended to type and automatically correcting any fat finger mistakes we may make without us taking pause or correcting words ourselves. I have learned to be slower on keys such as the “space bar” and reach further out to hit the letter more directly, but I shouldn’t have to make these adjustments. Regardless, I still can’t type as fast as I could on an iPhone.

Smaller Nits

• Browser double tap and zoom – When you double tap on a content area within a web page from the browser, the zoom often crops the text in a such a way that you have to scroll left and right and misses the mark. The Safari browser elegantly determines the content area you’re trying to focus in on and accurately zooms in without trying to reformat it.
• Notification count on app – The notifications window is a neat idea to be able to see everything and not have everything distract you with a pop-up as it does on the iPhone, but the implementation misses two critical features. The first is that I need to be able to see the notification and choose to not make it go away. For example, if I accidently touch the item it clears that notification, even if I haven’t read the SMS. Second, nothing on the application icon itself tells me that there is something needing my attention.
• SMS app still alerts me when I’m in it? That’s just silly.
• How did we go backward in basic PDA Functions? – The calendar application is missing the description field so if I click into an entry to know what the agenda is prior to walking into the meeting, I don’t’ have access. (Calendar missing description http://www.youtube.com/watch?v=eeTdJSzdVfQ). Task managed in Microsoft Outlook are not sync’d.
• Flick scrolling selection wheels –I’m sure this is one of those items that is simply missed as Apple has the patent, but it’s really annoying that I can’t just flick the hours forward or backward in the clock application to change my alarm. This simple slot machine type interface is so much easier than pushing a “+” or “-“ button.
• Passcode unlock – Why do I have to click the check mark?
• Using the physical key to get into application menu’s feels wrong (Droid X specific)
• Favorites in Phone (Contacts) can’t be sorted?
• Whoa battery life – Everyone gives the iPhone a hard time for not having “multi-tasking”, but other than listening to Pandora (now possible with iOS 4.0 API’s) I’d challenge anyone to show me that they can get in and out of things faster on a Droid. All I seem to get is applications that stay open and drain my batter. Yes, I realize I can get the Kill Apps app, but this should be more elegantly handled by the operating system (hear Fryo solves).
• iTunes integration – My life is in iTunes and iPhoto. Not being able to connect to these easily is very annoying (I blame you apple). So, I will eventually switch my photo album to Adobe’s Lightroom, which gives much better control of where the photos are kept as well as more capability, but I still haven’t figured out what to do with my music.
• Lack of apps – The first night I sat back on my couch with both my iPhone and new Droid X, the first thing I did was referenced my iPhone and opened the Android Marketplace to add all my apps…most were not there.
• Android Marketplace needs sub-categories

What I like

• The browser doesn’t refresh constantly when I return to it or hit the back button.
• Snooze is great on calendar notifications, but I should be able to set the snooze time
• Calendar widget is great at a glance
• Calendar allows me to decline a meeting later that I already accepted and send a note with it (iPhone just sent the decline without allowing me to add a message)

Overall it feels like an immature operating system that is being pushed out haphazardly and has had very little attention to the customer experience. That said, it’s young and has a lot of supporters so probably by the time this gets posted to my blog, much of my complaints will be solved. More importantly it’s an open platform that can be expanded on much more than the iPhone so I expect great things. Oh, and it’s nice to be able to watch hulu.com.

Introduction

Cars can be plugged in at the mechanics for electronic diagnosis, customer issues logged in enterprise support systems receive immediate potential solutions to their issue prior to a customer service representative looking at it, and computers send error reports when an application crashes. In industries across the world automated diagnostics becomes more and more prevalent leveraging continually advancing algorithms that become increasingly intelligent in identifying solutions to known problems. Yet in the health care industry Doctors have out dated and limited access to potential solutions and details from a patient’s case are seldom fully available to be investigated holistically.
There has been significant research on automated diagnosis, but limited practical application and integration of systems. The idea, well represented in Magnus Stensmo’s Ph.D thesis, Adaptive Automated Diagnosis, clearly paints the picture of how powerful and needed such an approach is.

Bias in today’s computer assisted diagnosis

“Enter symptom, disease type, test name or code” requests one physician diagnosis database. As with any human search that begins with keywords chosen by the user, bias inherently influences the results. If a Doctor has an assumed diagnosis, they will immediately begin searching for further evidence that their assumption can be validated. And if it isn’t, then they will have missed other potential diagnoses. Additionally, if the Doctor begins searching by symptoms, while these may be accurate, the order or weight given to any one symptom will give a bias toward related diagnosis when in fact, there may be a symptom not given any credit and thus not included in the search. Regardless of whether you consider today’s databases or the older process of researching in books, the results are always influenced by the bias of the researchers’ initial assumptions.

An Automated Medical Diagnosis System

What is needed instead is an approach that minimizes human bias and considers all relevant and irrelevant data in determining a diagnosis. Computer software does this well. With an automated medical diagnosis system, Doctors could be presented with multiple potential diagnoses based on all of the patient’s current and past details. Such a system could be designed for automated medical diagnosis that is based on probability, utility and decision theory (Read Adaptive Automated Diagnosis). Essentially, the computer software could be fed human observations of symptoms, test results, and any machine data collected such as blood pressure, heart rate, oxygen levels, etc. The software would then compare these observations with a database of potential diseases and external agents (e.g. , viruses, bacteria) to determine the most probable diagnosis.
These results would then be presented back to the doctor along with a probability rating indicating which ones are likely most relevant or accurate. Each diagnosis could also then be presented with additional direction to the doctor to further explore for additional symptoms and/or order an additional test. These additional observations and/or test results would again then be fed into the system where it could reevaluate the probable diagnoses canceling out some while raising the probability of others.
In addition to immediate interactions with the software, Intensive Care Unit’s machine observation data (e.g., oxygen levels, heart monitors) could be constantly fed into the system to allow the software to be looking for patterns that match other known diagnosis that would never be able to be caught by a human as it would take too much time to evaluate the data. Nurse’s notes could also be used in a similar manner.
The most challenging data for the automated medical diagnosis software to interrupt would be results of imaging systems. However, current advances in face recognition technology can and have been applied to reviewing images such as x-rays. This is important as “image interpretation is an error prone task. The number of lawsuits filed against medical imaging professionals that are related to the miss of a diagnosis is close to 70% (Berlin, 1995). The most common errors are perceptual errors that lead to diagnoses misses, representing about 60% of the cases (Renfrew et al., 1992).” – Application on Reinforcement Learning for Diagnosis Based on Medical Image . With a software solution, images could be reviewed against known patterns and then presented to a Doctor for a final review. Similarly, following a codification process, a standard could be defined to notate what one sees in an image that would be relevant to a medical diagnosis, which could then be understood by the software.
The database of potential diagnoses should be an online service that all medical institutions interact with in a real-time basis. This would allow for two key additional benefits. The first would be to connect Doctors in real time with their peers at other facilities with patients experiencing a similar condition. This would allow for immediate collaboration that could lead to a faster treatment. And second, by having the system online, you ensure that every doctor has access to the latest scientific diagnoses.
The system should then be designed to become increasingly more intelligent with time from increased information about each disease and ailment as well as feedback as to the accuracy of its results. Specifically, after each diagnosis, the Doctor would be responsible for submitting feedback as to whether or not the chosen diagnosis and subsequent treatment successfully resolved the patient’s problem. These responses could then go through an automated peer review with the results updating the probability factors to each disease and associated symptoms, test results, etc. Additionally, the software would record the other data learned from each patient in consideration of relevance to future cases. Results could be rated in accuracy based on how they are determined. For example, the results obtained from a biopsy or autopsy may be weighted as being more accurate than simply an observation from the Doctor that the patient recovers.

Available today

Today there are a number of point-of-care (POC) testing solutions, (also called near patient or bedside testing), but no large scale application of an automated medical diagnosis system. Today’s applications are for very focused blood tests and are far from providing the capabilities described above. Some examples:

• http://www.icutracker.net/
• http://www.rals.com/

The machine data required for the software has been developed such that it could be fed into such a system. There are many systems available today that would likely only require some standards to be developed for the resulting data. Some examples:

• Life support systems
• Positive patient ID
• Meds tracking
• Lab work tracking
• (see http://www.handheld.com/lp/healthcare.htm)

Conclusion

Software engineers everywhere can read this idea and will instantly recognize that today’s technology supports this solution. However, the greatest obstacles would be to gain public support for the requirements of fully electronic medical records and Doctors learning to work together. Progress is being made on changing the culture on both of these topics; unfortunately we have a long ways to go.
This article is dedicated to Grace Allen, a beloved mother, grandmother, great-grandmother, friend, cousin, partner and overall amazing woman. Had this system been in existence today, we would have had some more wonderful years with her.

References

• Stensmo, M & Sejnowski, T. J (1996) Automated Medical Diagnosis based on Decision Theory and Learning from Cases, World Congress on Neural Networks pp. 1227-1231
• Stensmo, Magnus (1995). Adaptive Automated Diagnosis. PhD thesis, KungligaTekniska Hogskolan (Royal Institute of Technology), Stockholm, Sweden.
• Stelmo Magalhaes Barros Netto, Vanessa Rodrigues Coelho Leite, Aristofanes Correa Silva, Anselmo Cardoso de Paiva and Areolino de Almeida Neto (2008). Reinforcement Learning ISBN 978-3-902613-14-1. Application on Reinforcement Learning for Diagnosis Based on Medical Image.
• Point of Care Testing
• http://www.pointofcare.net/
• http://www.meditech.com

Waiting for Windows to boot provides plenty of time for random contemplation. During a recent reboot, while stuck in small airplane seat, my hands began tapping away at the keyboard to mimic drum beats I’ve tapped out on many inanimate objects over the years (played the drums in high school). Lightly tapping the keys and hitting them hard produced different sounds that entertained me just enough for the boot. But while tapping away it dawned on me that the current Garage Band type keyboard setups for musical instruments are too limited when it comes to actually “playing” music on a QWERTY. Why can’t it use all the keys and associate groupings to one single drum head or key of an instrument? Why should I have to hit a single key to produce a single note?
A few years ago Palm released the updated Treo. The biggest part of its early fanfare was the keyboard. Their engineering led them to develop a smart keyboard that would know what key you meant to hit, even though you likely fat fingered a few keys because one key ultimately would be selected first in the I/O burst and thus could be assumed to be the intended key, allowing the smartphone to ignore the rest until the next major keystroke. Well, why not use the same approach to musical instruments on QWERTY’s? I’d like it if the bass drum was the combination of the space key, N, M, “,”, “.”, alt and some other funny windows menu key (HP Compaq 6910p). The snare drum could then be the lower left section containing keys such as the Z,X,C,S,D lot.
This could make for some fun jam sessions in the home office. You could also write the software to assume if you did hit more than one key, you were hitting harder.
I’m not sure what the computer gets for info from a keyboard, but if the I/O data is there, this would be a great enhancement to Garage Band. I also look forward to the iPhone application that allows you to tap out your music.

Many enterprise IT shops these days are significantly limiting mailbox sizes for their employees. It seems rather odd given the cost of storage, but none the less something many of us deal with. It means we’re constantly killing time each day cleaning up attachments and archiving mail. I shared my displeasure with a colleague of mine, James K. and he quickly created a Microsoft Outlook macro to automatically remove attachments. This feature should be a part of Outlook, but in the mean time, it saves me from opening each email individually to remove attachments (Outlook has no bulk processing feature for this).
Below is the macro that you can add to your own Outlook application:
——————————————————————-
Sub RemoveAttachments()
Dim selectedMailItem As Outlook.MailItem
Dim currentAttachment As Outlook.Attachment
Dim i As Integer
For Each selectedMailItem In ThisOutlookSession.ActiveExplorer.Selection
‘Remove attachments until there are none left
While selectedMailItem.Attachments.Count > 0
selectedMailItem.Attachments.Remove (1)
Wend
selectedMailItem.Save
Next
End Sub
——————————————————————-
Updated 9/24/07: Remove Attachments from Appointments
Sub RemoveAppointmentAttachments()
Dim selectedAppointmentItem As Outlook.AppointmentItem
Dim currentAttachment As Outlook.Attachment
Dim i As Integer
For Each selectedAppointmentItem In ThisOutlookSession.ActiveExplorer.Selection
‘Remove attachments until there are none left
While selectedAppointmentItem.Attachments.Count > 0
selectedAppointmentItem.Attachments.Remove (1)
Wend
selectedAppointmentItem.Save
Next
End Sub

This is just a short entry, but I keep meaning to write it. As I write this, I’m in the middle of removing attachments from emails in my sent folder in Microsoft Outlook. This is a huge pain. I have to open each one, right click on each attachment, and then select remove, close it, save it…
It should be as simple as me highlighting a number of emails all at once, then clicking something to remove all attachments. Done. Then I could manage my mailbox size better.

Openwave today announced their Contextual Merchandising Solution to Help Operators Monetize Off-Net Traffic. This system “enables the delivery of personalized content and advertising that is recommended to subscribers based on their user context and profile… [by] promot[ing] …complementary content and services based on a current usage patterns.” Every where around me I find companies getting better at helping me identify products that I would be most interested in. It began most obviously with Amazon, but has quickly evolved and many other online retailers take advantage of this capability such as Netflix (one of my personal favorites). However, I think the real opportunity is in taking the technology beyond the basic CRM data insights garnered from usage patterns and connecting it with insights from ones personal network and their “window shopping” patterns.
Let me start by explaining my thinking on personal network recommendations. In this context the system would be able to not only understand the individual’s buyer patterns, but also those of their friends. Through input from the user, the system would gain intelligence on which friend the individual user most aligns with on a given product category. The power of the personal network for purchase decisions is one that we should be able to bring to the mobile experience. An example would be someone shopping for a new digital camera and making note of their favorite one when out shopping, maybe even communicating they bought it. Then, when another friend is out shopping for a digital camera (or online) the system could let them know what ones their friends liked and purchased. Another example is picking out a hotel in a city you haven’t ever visited before. Wouldn’t it be nice to know if one of your friends stayed somewhere in that city and loved it?!
The second key input I think that would add valuable feedback to the user, would be to leverage their “window shopping” buying patterns. Imagine the insights that could be extracted from the foot path of a customer. How long do they pause in one area, where do they start, where do they abandon the store, what did they miss? To leverage this information you would use location based services to identify their foot path and then overlay it with a store map. You could also figure out where they went next (might be your competitor). With this information you could then redesign the layout of your store. Or, even more valuable to you and the customer, you could communicate directly to them when they’ve missed something you know they’d like as they walk out the door. Now, there are all kinds of considerations around privacy and the “creepiness” factor in this, but remember this, when American Express first launched computer telephony integration (CTI) with their interactive voice response (IVR) system to be able to answer the phone and say, “Hello Mr. Archer, how can we help you today?” people freaked! Today, caller ID is commonplace and expected (so frustrating when you enter your account info and the customer service rep asks for it again).
High end retailers (think “Nordstrom’s experience”) and big-box stores could dramatically increase in store sales if they leveraged such a capability. It could be a key differentiator…until everyone does it.

I believe I have gotten spoiled by the database. I use databases so often in my every day life that I just expect to be able to slice and dice data any way I want it. Sort it and filter it however, and most importantly only look at what I want to look at. Imagine for a moment if all Citysearch could do is show you every restaurant in their database all at once and had no idea to minimize the list by cuisine or neighborhood. Imagine if your cell phone only had one menu layer and everything was listed. Your life would be all about scrolling. Thankfully, we don’t have to deal with that…except for on my Comcast cable set-top box guide. Argh!
Here’s a thought. I’d like to see Comcast roll out a new on-screen guide that would allow me to set a preference to only show channels that I am subscribed to. I’m to the point where I feel like I’m chasing a moving target. Every time I sit down to watch something I choose a channel that says, “Subscription required…” I’m pretty sure my setup even changes what channels I have access to on a monthly basis to just keep me guessing. It’s very annoying and I usually give up. (Thank god for Netflix) In addition to this filter, I’d also like to see the ability to customize my menu. For example, does everyone have their local network’s FCC call letters memorized? Can’t they just put NBC on the menu?
Just a thought. Hey Comcast, give me a holler if you want to sit down and draft up some new requirements. 🙂