摘要：AGI 是指被认为与人类智力相当，甚至可能匹敌人类智力的人工智能。ASI 是指超越人类智力的人工智能，即使不是所有可行的方法，也在许多方面都优于人类。ASI 的理念是，ASI 能够在各个方面超越人类，从而超越人类。

本文来源：科技世代千高原

目前，大量研究正在进行，以进一步推进人工智能的发展。总体目标是实现通用人工智能（AGI），甚至可能实现超级人工智能（ASI）。

AGI 是指被认为与人类智力相当，甚至可能匹敌人类智力的人工智能。ASI 是指超越人类智力的人工智能，即使不是所有可行的方法，也在许多方面都优于人类。ASI 的理念是，ASI 能够在各个方面超越人类，从而超越人类。

我们尚未达到 AGI。

事实上，我们能否实现 AGI，或者说 AGI 是否能在几十年甚至几个世纪后实现，都还是未知数。目前流传的 AGI 实现日期众说纷纭，且缺乏任何可靠的证据或铁证如山的逻辑支撑。与传统人工智能目前的水平相比，ASI 更是遥不可及。

关于实现 AGI 的稻草人日期

由于在短期内实现 AGI 的可能性似乎比实现 ASI 更大，因此，让我们尝试预测如何实现 AGI。我将使用一些稻草人日期来帮助阐明这个模糊的问题。

最近对人工智能专家的调查显示，普遍的猜测是，2040年将是通用人工智能（AGI）实现的日期。许多人工智能领域的大佬们宣称我们将更早实现通用人工智能（AGI），比如从今天起的3到5年内，因此他们厚颜无耻地宣称将在2028年到2030年实现。但我认为这令人怀疑。他们甚至运用绝地武士的思维技巧，将通用人工智能（AGI）的定义扭曲得远低于其真正的含义，以此来支持他们大胆的日期预测。

我们这里使用的“稻草人”假设是2040年。这意味着我们有15年的跑道。思考一下这15年将如何发展是有益的。

时间线考虑

众所周知，我们目前正处于2025年的中期。试图设想在2040年实现通用人工智能（AGI）似乎是一项艰巨的任务。这与我们目前的人工智能状态相比，还有很长的路要走。

不用担心，我们将采取分而治之的方法，看看能找到什么办法。

一种可能性是，人工智能的进步逐年平稳推进，最终达到通用人工智能 (AGI)。假设每年都有渐进式的进步，且每年的进步幅度大致相同。换句话说，如果我们每年将人工智能的进步速度提高约 7%，持续大约 15 年，那么到 2040 年，通用人工智能 (AGI) 将成为现实（为了便于思考，我使用了四舍五入的数字）。

一些人工智能预言家认为，仅仅逐年提升人工智能水平并非通往成功的钥匙。他们认为，目前的方法和实践无法规模化发展。令人担忧的是，人工智能领域的每个人都陷入了一种“一刀切”的思维模式，盲目地追求同一种算法和方法。只有我们摆脱这种困境，提出全新的理念，通用人工智能 (AGI) 才有可能实现。想了解更多关于这场关于人工智能发展激烈争论的信息，请参阅我的报道（链接在此）。

奇迹的赌注

以下是那些对渐进式方法持强烈批评态度的人认为未来可能发生的情况。他们寄希望于这样一种想法：一位富有进取心的人工智能开发者能够奇迹般地突破现有人工智能的界限，并开发出一种前所未有的突破性新方法。这一突破将成为我们迈向通用人工智能 (AGI) 的“圣杯”。在发明或开发出这项不可思议的创新成果后不久，通用人工智能 (AGI) 将指日可待。

考虑一下这如何为时间线提供不同的视角。

也许这种渐进式的方法还能勉强维持十几年。虽然取得了一些进展，人们也不断沾沾自喜。但通用人工智能（AGI）似乎遥不可及。人工智能领域的投资者们开始感到不安，并开始思考通用人工智能何时才能真正实现。

突然间，这位富有进取心的人工智能开发人员在第 13 或 14 年左右取得了令人难以置信的突破。然后，这一突破被迅速培育成 AGI。

在这种情况下，人工智能会经历十二年的缓慢渐进式发展，然后突然被一种新的设计方式打断。一旦这种情况发生，人们所吹捧的通用人工智能（AGI）就会在相对较短的时间内被解决。这条时间线的变化大致相同，因为在这十五年里，人工智能会突然出现一种变革性的顿悟，让通用人工智能成为现实。这也许发生在第十年，而不是第十三年。又或许它发生在最后一刻，在第十四年出现。

这条时间线令人不安的问题是，它押注的是AGI追求过程中会发生某种奇迹。你可能看过一幅很受欢迎的漫画，画中两位科学家站在一块写满晦涩方程式的黑板前，黑板中间有一个明显的空隙。一位科学家问另一位，空隙里是什么？答案是，那里发生了奇迹。

七大途径

我提出了人工智能发展成为通用人工智能的七条主要路径。第一条路径是渐进式发展路径。人工智能行业倾向于将其称为线性路径。它本质上是缓慢而稳定的。“突然发生奇迹”的概念通常被称为“登月路径”。除了这两条路径之外，还有另外五条路径。

以下是我列出的从当代人工智能到珍贵的通用人工智能的七条主要途径：

你可以将这七种可能的路径应用到任何你想设想的时间线中。我以2040年实现AGI的15年为例进行说明。2050年更有可能实现，而且这将持续25年。如果2028年是AGI的到来年，那么这条路径将会被显著压缩。

下注

对某条路径而非另一条路径的信念如何影响你的下注？

如果线性路径是你放置筹码的地方，那么似乎只需要继续做现在正在做的事情。保持船身稳定，并保持在正确的航向。不要让任何事情分散你的注意力。

通过“登月计划”突然跃升至通用人工智能 (AGI)，似乎需要与当前的做法截然不同。尽一切可能打破常规，为那些天马行空、充满想象力的新想法提供资金。培育它们，不要让别人的短视压力改变你的方向。

类似的策略适用于各个相应的途径。

我敢打赌，你一定非常好奇，这七条路径中哪一条被认为最有可能。此外，你或许也对这七条路径中哪一条被认为最不可能略感兴趣。

在与许多人工智能研究人员同行的交流中，我有一种非常随意且非正式的感觉：S曲线是最有可能的。这通常与高科技发展曲线相符。它也遵循着这样的信念：我们现在所做的事情不会规模化。在平台期，一些新的变化将推动我们前进，打开规模化的大门。这不会是奇迹般的突破。相反，独创性和创新性将有助于推动发展。

七条路径中哪一条符合你的心意？

至于最不可能的途径，人工智能领域的同事们也同样临时抱佛脚地推测，登月计划不会成为我们迈向通用人工智能的救星。在他们看来，这种灵丹妙药的概率甚至比流星落在你头上时闪电击中你还要低。或许这种怀疑态度反映出一种信念：我们所知道的就是我们所知道的，没有其他我们尚未发明的非凡之物。

我当然不希望这种情绪阻碍任何人工智能创新者突破界限、尝试宏大的新想法。请保持坚强的精神。不要让反对者阻碍你追寻内心的追求。

正如著名的美国艺术史学家所说：“奇迹会发生在那些相信奇迹的人身上。”获得 AGI 也可能会发生同样的事情。

Big Bets On Which Of These Pathways Will Push Today’s AI To Become Prized AGI
ByLance Eliot, Contributor. Dr. Lance B. Eliot is a world-renowned AI scientist and consultant.

Follow Author
May 04, 2025, 03:15am EDT

Share

Save

Comment


Laying out the pathways from today's convention AI to reach the vaunted AGI.
GETTY
In today’s column, I examine the most likely pathways to get us from today’s contemporary AI to the vaunted AGI (artificial general intelligence). This is a mighty big open question and there are AI makers and humongous tech firms all making bets on which path will be the winner-winner chicken dinner when it comes to attaining AGI.

Let’s talk about it.

This analysis of an innovative AI breakthrough is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).

Heading Toward AGI And ASI
First, some fundamentals are required to set the stage for this weighty discussion.

PROMOTED

There is a great deal of research going on to further advance AI. The general goal is to either reach artificial general intelligence (AGI) or maybe even the outstretched possibility of achieving artificial superintelligence (ASI).
AGI is AI that is considered on par with human intellect and can seemingly match our intelligence. ASI is AI that has gone beyond human intellect and would be superior in many if not all feasible ways. The idea is that ASI would be able to run circles around humans by outthinking us at every turn. For more details on the nature of conventional AI versus AGI and ASI, see my analysis at the link here.

We have not yet attained AGI.
In fact, it is unknown as to whether we will reach AGI, or that maybe AGI will be achievable in decades or perhaps centuries from now. The AGI attainment dates that are floating around are wildly varying and wildly unsubstantiated by any credible evidence or ironclad logic. ASI is even more beyond the pale when it comes to where we are currently with conventional AI.
Strawman Dates On Attaining AGI
Since attaining AGI seems to be a greater chance in the relatively near-term versus achieving ASI, let’s put our minds to trying to foresee how AGI is going to be reached. I will use some strawman dates to help illuminate the murky matter.

Recent surveys of AI specialists indicate that the overall guess is that the year 2040 will be the presumed date by which AGI will have been accomplished. Numerous AI luminaries are touting that we will arrive at AGI sooner, such as in the next 3 to 5 years from today, thus they are staking their brazen claims on the years 2028 to 2030. I find this to be doubtful. They are also using Jedi mind tricks to twist the definition of AGI into being a lot less than what AGI is really supposed to denote, which helps to bolster their emboldened date forecasts. For my analysis of the various predicted dates and assorted definitions of AGI, see the link here.

The strawman we will use here is the year 2040. That gives us a runway of 15 years. It is useful to put some thought into how those fifteen years are going to play out.
Timeline Considerations
As you well know, we are currently sitting just about mid-way through the year 2025. Trying to envision arriving at AGI in the year 2040 seems like a daunting task. It is quite a long distance in time from our present-day AI status.
No worries, we will do a divide-and-conquer approach to see what we can come up with.
One possibility is that the advances in AI occur smoothly on a year-by-year basis which ultimately culminates in AGI. Assume that each year there is an incremental advancement, and the advancement is roughly the same amount of progression each year. In other words, if we improve AI by about 7% per year, doing so over roughly 15 years, AGI will become a reality by 2040 (I’m using rounded numbers for this thought exercise).
Some AI prognosticators believe that simply incrementing AI each year is not the ticket to success. Their view is that the current methodologies and practices are not going to scale up. Concerns are that everyone in AI is pretty much part of a massive one-way-fits-all mindset, blindly pursuing the same kinds of algorithms and approaches. Only if we break free of this malaise and come up with radically new ideas will AGI be attained. For more on this AI progression heated debate, see my coverage at the link here.
The Bet On A Miracle
Here’s what vocal critics of the incremental approach say is potentially going to happen. Their hope is pinned on the idea that an enterprising AI developer will miraculously see beyond the bounds of existing AI and derive a groundbreaking new approach that no one has yet even imagined. This breakthrough will be the Holy Grail that gets us to AGI. Shortly after inventing or figuring out this incredible innovation, AGI will be right around the corner.
Consider how this gives a different perspective on the timeline.
Maybe the incremental approach muddles along for a dozen years. Some progress is being made, and ongoing self-congratulations are occurring. But AGI doesn’t seem within view. Investors in this AI are getting perturbed and asking hard questions about when AGI is finally going to be had.
Boom, out of nowhere, the enterprising AI developer comes up with an incredible breakthrough, doing so around year 13 or 14. Then, this breakthrough is rapidly nursed into becoming AGI.
In that scenario, there are twelve years of modest incremental progress that is then suddenly punctuated by a new way of devising AI. Once that occurs, in relatively short order the vaunted AGI is figured out. Variations on that timeline are roughly the same in the sense that over the fifteen years, there is a sudden transformative eureka about AI that puts AGI in the picture. Perhaps this happens in year 10 instead of year 13. Or maybe it occurs at the last moment, arising in year 14.
A disconcerting problem with that timeline is that it is a bet on a kind of miracle occurring during the AGI pursuit. You might have seen a popular cartoon of two scientists standing at a chalkboard that is filled with arcane equations, and in the middle of the chalkboard, there is a noticeable gap. One scientist asks the other one, what goes in that gap? The response is that a miracle goes in that spot.
Seven Major Pathways
I’ve come up with seven major pathways that AI is going to advance to become AGI. The first listed path is the incremental progression trail. The AI industry tends to refer to this as the linear path. It is essentially slow and steady. The idea of a sudden miracle happening is usually coined as the moonshot path. Besides those two avenues, there are five more.
Here’s my list of all seven major pathways getting us from contemporary AI to the treasured AGI:
* (1) Linear path (slow-and-steady): This AGI path captures the gradualist view, whereby AI advancement accumulates a step at a time via scaling, engineering, and iteration, ultimately arriving at AGI.
* (2) S-curve path (plateau and resurgence): This AGI path reflects historical trends in the advancement of AI (e.g., early AI winters), and allows for leveling-up via breakthroughs after stagnation.
* (3) Hockey stick path (slow start, then rapid growth): This AGI path emphasizes the impact of a momentous key inflection point that reimagines and redirects AI advancements, possibly arising via theorized emergent capabilities of AI.
* (4) Rambling path (erratic fluctuations): This AGI path accounts for heightened uncertainty in advancing AI, including overhype-disillusionment cycles, and could also be punctuated by externally impactful disruptions (technical, political, social).
* (5) Moonshot path (sudden leap): Encompasses a radical and unanticipated discontinuity in the advancement of AI, such as the famed envisioned intelligence explosion or similar grand convergence that spontaneously and nearly instantaneously arrives at AGI (for my in-depth discussion on the intelligence explosion, see the link here).
* (6) Never-ending path (perpetual muddling): This represents the harshly skeptical view that AGI may be unreachable by humankind, but we keep trying anyway, plugging away with an enduring hope and belief that AGI is around the next corner.
* (7) Dead-end path (AGI can’t seem to be attained): This indicates that there is a chance that humans might arrive at a dead-end in the pursuit of AGI, which might be a temporary impasse or could be a permanent one such that AGI will never be attained no matter what we do.
You can apply those seven possible pathways to whatever timeline you want to come up with. I used the fifteen years of reaching AGI in 2040 as an illustrative example. It could be that 2050 is more likely and this will play out over 25 years. If 2028 is going to be the AGI arrival year, the pathway is going to be markedly compressed.
Placing Your Bets
How does a belief in one pathway over another pathway shape the placing of your bets?
If the linear path is where you are putting your poker chips, it would seem that all that needs to happen is to continue doing what is already being done right now. Keep the ship steady and presumably on course. Don’t let anything distract from that direction.
The sudden leap to AGI via the moonshot path would appear to necessitate a maverick departure from what is being done at this time. Do whatever is feasible to think outside the prevailing box. Fund those wild and wide-eyed new ideas. Nurture them along and do not let the myopic pressures of others convince you otherwise.
Similar strategies apply to each respective pathway.
I’m betting you are avidly curious as to which of the seven pathways is thought to be the most likely. In addition, you might be mildly interested in which of the seven is seen as the least likely.
In talking with many of my fellow AI researchers, a casual and highly informal sense is that the S-curve is the most likely. This generally aligns with high-tech development curves. It also abides by the belief that what we are doing now isn’t going to scale up. During a period of a plateau, some new change is going to nudge us forward and open the door to scaling up. It won’t be a miracle breakthrough. Instead, ingenuity and novelty will help move the needle.
Which of the seven pathways suits your fancy?
In terms of the least likely of the pathways, the same ad hoc semblance of AI colleagues speculates that the moonshot won’t be the rescuer to get us to AGI. In their minds, the miracle cure gets worse odds than lighting striking you while a meteor lands on your head. Maybe this skepticism reflects a belief that what we know is what we know and that there isn’t something else extraordinary that we haven’t yet devised.
I certainly don’t want that sentiment to dampen any AI innovators from stretching boundaries and trying outsized new ideas. Please keep your spirit strong. Do not let naysayers stop you from your heart’s pursuit.
As the famous American art historian remarked: “Miracles happen to those who believe in them.” The same might happen with attaining AGI.

来源：人工智能学家