The State of Education Video

Video’s role in schools is taken for granted entering 2023, although we should expect to see changes, potentially disruptive, in the educational video market as schools continue to adapt to the aftermath of the COVID-19 state of emergency phase. Despite the widely held belief that video is essential to school operations, expect to see schools roll back their investments in video services, while educators seek out ways to go beyond the basics of video delivery, finding better ways to engage students both with synchronous and asynchronous video.

Time for De-Escalation?

It stands to reason that at some point, educational institutions will de-escalate their investments in educational video platforms from the peak seen during widespread school closures of 2020 and into 2021. With at least a year of in-person school operations behind us, 2023 is likely to be when we start to see that drop-off along with a decline to a new-normal adoption rate for video platforms. Since Kaltura, the largest provider of educational video platform services, completed its IPO in July 2021, we have an excellent and reliable resource in its periodic SEC filings to gain insight into our industry’s trends.

While Kaltura’s reported subscription revenue (Figure 1) has increased in each of its last three quarterly reports as compared to the same quarters in 2021, the rate of change shows a decreasing trend. There are confounding factors in this dataset: Kaltura revenue for this segment is typically collected in annual payments, but it is a fair interpretation that this supports a trend toward reaching a peak at some point in 2023, followed by decline to a new steady growth state similar to what was seen pre-pandemic. This de-escalation is predicted broadly across tech: In late 2022 and into 2023, almost every major tech company, including Alphabet/Google, Meta/Facebook, Microsoft, and even Kaltura, has deemed it necessary to enact historically significant layoffs in this transition as the COVID crisis ebbs.

Figure 1. Kaltura’s enterprise, education, and technology subscription revenue, with a chart automatically generated by Excalidraw, an amazing, free E2E encrypted whiteboard collaboration tool that I recommend for instruction and meetings.

Before moving on from reviewing the movement among the industry’s major players, let’s talk about perhaps the most unexpected development in 2022: Panopto’s attempts to acquire Kaltura. In July, Panopto (backed by K1 Investment Management) made an unsolicited offer to acquire 93.1% of Kaltura stock for $3 per share, at a total cost of around $383 million. (K1 already owns the remaining 6.9% of Kaltura.) While this was less than 30% of the IPO valuation from the previous summer, Kaltura stock at the time had been selling for under $2 per share from March, compared to the IPO pricing of $10 per share, so the offer may have been tempting to shareholders.

On Sept. 2, the Kaltura board declined the offer, and the drama concluded. Had the deal gone through, it would have been a hugely consequential consolidation of the two largest players in the educational video platform-as-a-service (PaaS) market. The Kaltura and Panopto platforms are very different in design and underlying technologies, aside from both running in AWS clouds, so it would have been curious to see how the merger would have played out, given Panopto’s more aggressive entry into the enterprise video vertical since the K1 partnership and Kaltura’s expansion into the asynchronous market.

De-Escalation in Practice

Back to our main theme: For schools, what would this predicted de-escalation look like? Most likely, they would simply lower their usage allotment with their existing PaaS vendor to wind things down, dumping unused content and slowing the pace at which they add new material. This is a decision school administrators would be adequately informed to make. They know what their video needs were when their school was completely shut down, and they know what their resource needs were last year with their courses back in the physical classroom at some percentage between 0 and 100. Based on this information, they could adjust their subscription downward to the usage they expect to need.

With video established as an essential tool for schools, what would they do for video services if institutional decision makers elected to allow their contract with a video PaaS provider to lapse? They might opt for a “free” hosting platform with some policy guardrails to avoid privacy issues from creating litigable problems. Many schools would look into handling their video-hosting needs on-premise, using either the venerable and well-maintained Opencast (née Opencast Matterhorn) open source video platform, Kaltura’s community edition, or a proprietary video platform that they could run more hands-on than what they had paid a premium for in the PaaS market.

Somewhat surprisingly, Opencast is not packaged with an open source media server, with the presumption that schools will use either Wowza or YouTube as a video host. Smaller schools that are eager to bring video-hosting services back on-premise and are using open source software to minimize costs have two good options. The first is to host the video files statically by pre-fragmenting them and generating static M3U8 manifest files for HTTP Live Streaming delivery. The second option is to use the NGINX RTMP module (go2sm.com/nginx), which is maintained by an F5 employee (recall that F5 acquired NGINX back in 2019 for $670 million). Either approach would make it easy to restrict access to fragments and playlists of curricular video or video showing students using their school’s single-sign-on solution.

In addition, both options would be financially sound for an on-premise solution. In the cloud, variable-use resources like CPU, RAM, and bandwidth tend to be very cost-effective, while fixed-use resources like long-term disk storage are more expensive than what you can get with an on-premise investment. In other words, economies of scale work the best when you pay for what you’re using to serve customers and hand those resources off to other public cloud tenants at other times. They function the worst when you’re always paying to store data you’ve accumulated and may or may not be using.

Using NGINX with its RTMP module is exactly how Kaltura’s open source offering is deployed utilizing its all-in-one installers, but Kaltura also publishes its own robust MP4 repackager module for NGINX (go2sm.com/kalturadynamic) as well as a sophisticated Live Media Framework for live streaming (go2sm.com/livemedia). That Kaltura’s continued support for the open source version of its platform presents an attractive exit strategy for cash-strapped schools is not lost on Kaltura’s executive management. In fact, it was discussed as a significant threat to company revenues on pages 26 and 27 of its 2022 10-K filing.

In 2023, I expect that in-class synchronous video usage will decline for most schools as the transition away from emergency remote teaching takes hold, although synchronous video platform subscriptions are unlikely to lapse entirely. Video meetings remain essential to school operations; faculty members often prefer to have a videoconference instead of walking across campus for a routine meeting.

Despite this expectation, Zoom revenue growth as reported in its quarterly 10-Q filing shows a familiar quarterly trend (Figure 2). Even Zoom’s transcendence across constituencies is not immune to some degree of the post-COVID de-escalation.

Figure 2. Here is a comparable Zoom total revenue figure made with Excalidraw. Kaltura’s revenue had subscription revenue broken out from consulting services, where it saw actual dollar declines since subscribers were presumably already set up.

AI Gets Real

OpenAI made enormous waves among educators late in 2022, as numerous published articles predicted how its text-generation system, ChatGPT, would spell the end of writing as a tool to evaluate students (or prospective students). While ChatGPT shows remarkable capabilities, both in its ability to interpret user prompts and to generate text passages, it typically produces mediocre, lowest-common-denominator writing. That is indeed a threat, both to schools that are trying to assess student performance and to students and knowledge workers with developing skills whose writing could be mistaken for ChatGPT output. Text-generation engines like this will no doubt be abused to produce garbage text in classrooms and on the internet, much like more basic Markov chain text generators have for well over a decade, albeit theirs is higher-quality garbage.

All that is not to overdo the pooh-poohing. I too have seen impressive cases of people using ChatGPT as a search engine and getting correct answers. It’s not all useless garbage, but students do need to set their self-expectations higher than what ChatGPT can do. If ChatGPT’s current shortcomings prevent it from posing an immediate threat to the college and high school essay, it does better (although not especially well, depending on your stance on grade inflation) at interpreting short-answer question prompts and providing a correct short answer, so concerns may be more justified there.

But the more interesting and potentially transformative of OpenAI’s products made available recently is Whisper (go2sm.com/whisper), its speech-to-text (STT) system that was open sourced on Sept. 21, 2022. Whisper was trained on a corpus of 680,000 hours of multi-language speech data using “weakly” supervised machine learning methods. For an interesting example of this weak supervision, Whisper’s training set was automatically filtered of machine-generated transcripts purportedly aligned to an audio sample using the same sorts of tools that teachers will become familiar with to identify AI-generated text passages. Whisper’s accuracy is nothing short of remarkable, especially when compared to previous open source STT engines. I try to avoid using terms like “game-changer,” but Whisper is the real deal. Try the online demo (whisper-openai.vercel.app) for yourself (Figure 3). For evidence of how impressive Whisper is, Opencast 13.0, released 84 days after Whisper on Dec. 14, 2022, already offered it as an available STT engine for generating video captions, in addition to cloud STT engines from Amberscript, Microsoft, IBM, and Google or the open source Vosk engine.

Whisper

Figure 3. Whisper onloine demo

Including Whisper is another way that the Opencast project is approaching feature parity with the major vendor option video platforms. It also includes the Tesseract OCR engine for extracting metadata from slide contents shown in videos, which is potentially beneficial both for content search and seeding audio description tracks. Opencast development is largely financed by European universities and schools, where open source is looked on more favorably than in U.S. schools and where vendors are perceived as more reliable and cost-effective, as well as attractive for FOMO (fear of missing out) or interoperability motivations.

Also worth acknowledging here are Adobe’s major strides in incorporating STT into its product lines, notably with Premiere Pro, which is able to generate a quite accurate caption track for a sequence that can be edited along with the video. Adobe Podcast (formerly Project Shasta; see Figure 4) launched in 2022 to compete with Descript, a podcast authoring and editing application that I’ve praised in the past. It includes some impressive features that, if the zaniest hype were to be believed, will do to audio engineers what ChatGPT will do to technical writers.

Figure 4. Adobe Podcast (formerly Project Shasta)

Demanding More From Video Platforms

By now, the availability of video hosting for teaching and learning is something teachers take for granted, and they want to see more from a video platform than the basics. Specifically, they’re looking for ways to make their synchronous videos more engaging for students. Video quiz is the most explored type of interactive video, in which quiz questions can serve as prompts for what is to come, for self-assessments for what came before, or merely for milestones to enforce that the video was watched in its entirety. Video quiz is a major differentiator for PaaS vendors over Opencast or other alternatives, although it tends to be a minimalistically implemented feature in every full-fledged video platform I’ve seen. The leader of the pack for video quiz is PlayPosit (acquired by WeVideo in August 2022), which maintains its advantage on the strength of a well-designed question editor, a larger set of question types, and integration with the major video-hosting platforms (Figure 5).

Figure 5. Current PlayPosit integration with LMSs and video-hosting platforms

Lower-stakes forms of engagement are also popular in today’s educational environments. Among those available on the market are in-video chat and hotspots to direct students to remedial or “in case you’re interested” resources when a video lecture has to gloss over a topic that may be important or interest-piquing.

Another area of need for schools is verification of student work for online activities. If we fear that ChatGPT or a mercenary scholar-for-hire is likely to be performing work for students and that our schools’ reputations will be sullied by graduating unprepared students who are awarded unearned grades and degrees, we need to build guarantees into the course activities that the student registered is the one who did the evaluated work. Video offers obvious ways to provide those assurances, but student-generated content is a bit of a different beast than faculty-produced video. For students, the video-authoring interface needs to be as easy to operate as any other assignment-submission mechanism. The hosting of these videos also has to protect student privacy both from other students enrolled in the course as appropriate and from everyone outside the course. Finally, the hosting may need to abide by the school’s data-retention policies, which vary from school to school, including cases in which the student might contest their grade in the course years after taking it and would be entitled to use their video submission as evidence of how they were unfairly graded if the data-retention policy says that it should have been retained.

Free

for qualified subscribers

Subscribe Now Current Issue Past Issues